Text-to-speech plans compared
54 text-to-speech plans from 15 providers, checked 2026-11-16 against the official pages. 260 features compared, 240 read from official pages by our pipeline, 0 measured by us.
Price statistics for text to speech · No audio has been made yet, so the listening slots and measured figures below are empty and say so.
Choose what to compare
Listen to the same sentence
Same sentence, every plan: choose a language and a sentence, press play on each plan. Plans without a sample say so.
2026년 3월 15일 오후 2시 30분에 서울역 3번 출구에서 만나요.
- ElevenLabs Creator Speaks Korean Not yet available to listen to
- OpenAI tts-1-hd Speaks KoreanNot yet available to listen to
source
Kazakh, Korean, Latvian
developers.openai.com/api/docs/guides/text-to-spee · checked 2026-11-13 - Google Cloud Chirp 3: HD voices Speaks Korean Not yet available to listen to
- Amazon Web Services Neural voices Speaks Korean Not yet available to listen to
11 of 15 products name Korean on their official page
- Cartesia Text to Speech API Speaks Korean
- Deepgram Aura Text to Speech Speaks Korean
source
`flux-{voice}-ko` | Korean: `ko`
developers.deepgram.com/docs/tts-models-languages- · checked 2026-11-13 - ElevenLabs Text to Speech API Speaks Korean
- Fish Audio API Speaks Korean
source
French, Spanish, Korean, Arabic,
docs.fish.audio/developer-guide/models-pricing/mod · checked 2026-11-13 - Google Cloud Text-to-Speech Speaks Korean
- Gradium Text to Speech API Korean is not in its list
source
We currently support five languages: English (en), French (fr), German (de), Spanish (es) and Portuguese (pt)
docs.gradium.ai/guides/introduction.md · checked 2026-11-13 - Inworld Text to Speech Speaks Korean
- Lemonfox Text-to-Speech API Korean is not in its list
source
We support the following languages: en-us , en-gb , ja , zh , es , fr , hi , it , or pt-br
www.lemonfox.ai/apis/text-to-speech · checked 2026-11-13 - MiniMax Speech API Speaks Korean
- Murf API Korean not stated
source
turns your text into speech, with 150+ voices in 35 languages
murf.ai/api/docs/introduction/overview.md · checked 2026-11-13 - OpenAI Text to Speech Speaks Korean
source
Kazakh, Korean, Latvian
developers.openai.com/api/docs/guides/text-to-spee · checked 2026-11-13 - Amazon Polly Speaks Korean
- Rime Text to Speech API Korean is not in its list
source
it serves ten languages: English, Arabic, Dutch, French, German, Hindi, Italian, Japanese, Portuguese, and Spanish
docs.rime.ai/docs/introduction.md · checked 2026-11-13 - SpeechifyAI Text to Speech API Speaks Korean
- Typecast API Speaks Korean
source
ara, ben, bul, yue, zho, hrv, ces, dan, nld, eng, fin, fra, deu, ell, hin, hun, ind, ita, jpn, kor, msa, nan, nor, pol, por, pan, ron, rus, slk, spa, swe, tgl, tam, tha, tur, ukr, vie
typecast.ai/docs/llms.txt · checked 2026-11-13
Side by side
260 features compared, 240 read from official pages by our pipeline, 0 measured by us. Open a cell's source to see the official sentence, its page and the date we read it. A cell that says "not stated on the pages we read" is not a no.
Price and billing 3 rows
| Feature | ElevenLabs Creator | OpenAI tts-1-hd | Google Cloud Chirp 3: HD voices | Amazon Web Services Neural voices |
|---|---|---|---|---|
| Monthly cost at 1,000,000 characters | from $18.33; the rest is not stated on the pages we readhowNeeds 1,000,000 characters; the plan includes 275,000 and the extra-usage terms are not stated on the official page. · 2026-11-16 | $30.00how1,000,000 characters = 1,000,000 characters × $0.03 per 1,000 → $30/month. · 2026-11-16 | $0.00how1,000,000 characters = 1,000,000 characters × $0.03 per 1,000, minus the $30 monthly credit → $0/month. · 2026-11-16 | $16.00how1,000,000 characters = 1,000,000 characters × $0.016 per 1,000 → $16/month. · 2026-11-16 |
| Billing | annual subscription | pay as you go | pay as you go | pay as you go |
| Price per 1,000 characters | not stated on the pages we read | $0.03sourcetts-1-hd | Text | $30.00 / 1M characters | - | -developers.openai.com/api/docs/pricing.md · checked 2026-11-16 | $0.03sourceChirp 3: HD voices (sku:F977-2280-6F1B) 0 to 1 million characters US$0.00003 per charactercloud.google.com/text-to-speech/pricing · checked 2026-11-16 | $0.016sourceAmazon Polly’s Neural voices are priced at $16.00 per 1 million characters for speech or Speech Marks requested (when outside the free tier).aws.amazon.com/polly/pricing · checked 2026-11-16 |
Limits 1 rows
| Feature | ElevenLabs Creator | OpenAI tts-1-hd | Google Cloud Chirp 3: HD voices | Amazon Web Services Neural voices |
|---|---|---|---|---|
| Included each month | 275,000 characters | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
Features (judged by the engine) 48 rows
| Feature | ElevenLabs Creator | OpenAI tts-1-hd | Google Cloud Chirp 3: HD voices | Amazon Web Services Neural voices |
|---|---|---|---|---|
| Commercial use of the audio | Yessourceif you access or use our Services free of charge (such a user, a “ Free User ”), you may only use the Services for non-commercial purposes; (ii) if you access or use our Services through a paid subscription plan (such a user, a “ Paid User ”), you may use the Services for commercial purposeselevenlabs.io/terms-of-use · checked 2026-11-16 | not stated on the pages we read | YessourceYou can use the audio data files you create using Cloud Text-to-Speech in your applications or media in compliance with the Google Cloud Platform Terms of Servicecloud.google.com/text-to-speech/quotas · checked 2026-11-16 | YessourceAs between you and AWS, your Polly output belongs to you.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| Voice cloning | YessourceEverything in free, plus Commercial License Instant Voice Cloningelevenlabs.io/pricing/api · checked 2026-11-16 | not stated on the pages we readsourceCustom voices created from recordings are limited to eligible customers.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-16 | not stated on the pages we readsourceImportant: Access to Instant Custom Voice is restricted to allow-listed users.cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-16 | not stated on the pages we readsourceHow do I get started with Amazon Polly Brand Voice? If you are interested in building a Brand Voice using Amazon Polly, please reach out to your AWS Account Manager or contact us for more information.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| Streaming audio | YessourceThe Text-to-Speech WebSockets API is designed to generate audio from partial text input while ensuring consistency throughout the generated audio.elevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-16 | YessourceThis means the audio can be played before the full file is generated and made accessible.developers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-16 | YessourceStreaming Cloud Text-to-Speech is only compatible with Chirp 3: HD voicescloud.google.com/text-to-speech/docs/create-audio… · checked 2026-11-16 | not stated on the pages we read |
| Korean | Yes | YessourceHindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Kannada, Kazakh, Korean, Latvian, Lithuaniandevelopers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-16 | YessourceKorean (South Korea) Premium ko-KR ko-KR-Chirp3-HD-Achernar FEMALEcloud.google.com/text-to-speech/docs/list-voices-… · checked 2026-11-16 | YessourceJapanese ja-JP Korean ko-KR Norwegian nb-NOdocs.aws.amazon.com/polly/latest/dg/supported-lan… · checked 2026-11-16 |
| Your text is not used to train models | not stated on the pages we readsourceyou may opt out of our use of your Content for training at any time by navigating to the 'Data use' menu in the 'Terms and Privacy' section of your ElevenLabs account.elevenlabs.io/terms-of-use · checked 2026-11-16 | Yessourcedata sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us).developers.openai.com/api/docs/guides/your-data.md · checked 2026-11-16 | not stated on the pages we readsourceStreaming Cloud Text-to-Speech is only compatible with Chirp 3: HD voicescloud.google.com/text-to-speech/docs/create-audio… · checked 2026-11-16 | not stated on the pages we readsourceAmazon Polly may store and use text inputs processed by the service solely to provide and maintain the service and to improve and develop the quality of Amazon Polly and other Amazon machine-learning/artificial-intelligence technologies.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| Free allowance that repeats every month | not stated on the pages we read | not stated on the pages we read | YessourceChirp 3: HD voices (sku:F977-2280-6F1B) 0 to 1 million characters US$0.00003 per charactercloud.google.com/text-to-speech/pricing · checked 2026-11-16 | not stated on the pages we readsourceFor Neural voices, the free tier includes 1 million characters per month for speech or Speech Marks requests, for the first 12 months.aws.amazon.com/polly/pricing · checked 2026-11-16 |
| Official MCP server | YessourceThe ElevenLabs hosted MCP server is a remoteelevenlabs.io/docs/eleven-agents/operate/hosted-m… · checked 2026-11-16 | YessourceServer URL (streamable HTTP):** `https://developers.openai.com/mcp`developers.openai.com/learn/docs-mcp.md · checked 2026-11-16 | not stated on the pages we readsourceStreaming Cloud Text-to-Speech is only compatible with Chirp 3: HD voicescloud.google.com/text-to-speech/docs/create-audio… · checked 2026-11-16 | not stated on the pages we readsourceAs between you and AWS, your Polly output belongs to you.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| Remote MCP server | YessourceThe ElevenLabs hosted MCP server is a remoteelevenlabs.io/docs/eleven-agents/operate/hosted-m… · checked 2026-11-16 | YessourceServer URL (streamable HTTP):** `https://developers.openai.com/mcp`developers.openai.com/learn/docs-mcp.md · checked 2026-11-16 | not stated on the pages we readsourceStreaming Cloud Text-to-Speech is only compatible with Chirp 3: HD voicescloud.google.com/text-to-speech/docs/create-audio… · checked 2026-11-16 | not stated on the pages we readsourceAs between you and AWS, your Polly output belongs to you.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| MCP sign-in method stated | YessourceAuthentication uses OAuth. When you connect, you sign in with your ElevenLabs account and grant the assistant scoped access to your workspace.elevenlabs.io/docs/eleven-agents/operate/hosted-m… · checked 2026-11-16 | not stated on the pages we readsourceServer URL (streamable HTTP):** `https://developers.openai.com/mcp`developers.openai.com/learn/docs-mcp.md · checked 2026-11-16 | not stated on the pages we readsourceStreaming Cloud Text-to-Speech is only compatible with Chirp 3: HD voicescloud.google.com/text-to-speech/docs/create-audio… · checked 2026-11-16 | not stated on the pages we readsourceAs between you and AWS, your Polly output belongs to you.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| SOC 2 report | Yes | not stated on the pages we read | Yessourcefor standards such as ISO 27001, ISO 27017, ISO 27018, SOC 2, SOC 3, and PCI DSS.docs.cloud.google.com/document-ai/docs/security · checked 2026-11-16 | YessourceAWS SOC 2 Security, Availability, Confidentiality, & Privacy Report available to AWS customers from AWS Artifact .aws.amazon.com/compliance/soc-faqs · checked 2026-11-16 |
| Data region stated | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Data processing agreement | Yes | not stated on the pages we read | Yes | YessourceThese Service Terms incorporate the AWS Data Processing Addendum (“DPA”), when you use AWS Services to process Customer Data (as defined in the DPA).aws.amazon.com/service-terms · checked 2026-11-16 |
| Uptime SLA | not stated on the pages we read | not stated on the pages we read | Yessource(i) For GCP Services, each of the then-current service level agreements atcloud.google.com/terms · checked 2026-11-16 | not stated on the pages we read |
| Pronunciation dictionary | YessourceCreate a pronunciation dictionary from ruleselevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-16 | not stated on the pages we read | YessourceCustom pronunciations is available across all locales except: bn-IN , gu-IN ,cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-16 | YessourceYou can modify the pronunciation of particular words, such as company names, acronyms, foreign words and neologismsaws.amazon.com/polly/faqs · checked 2026-11-16 |
| Cancel at any time | Yes | not stated on the pages we read | not stated on the pages we read | YessourceYou may terminate this Agreement for any reason by providing us notice and closing your account for all Services for which we provide an account closing mechanism.aws.amazon.com/agreement · checked 2026-11-16 |
| You own the text you send | Yes | not stated on the pages we read | Yessourceproperty. As between the parties, Customer retains allcloud.google.com/terms · checked 2026-11-16 | YessourceYou always retain ownership of your content and we will only use your content with your consent.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| You own the generated audio | Yes | not stated on the pages we read | not stated on the pages we read | YessourceAs between you and AWS, your Polly output belongs to you.aws.amazon.com/polly/faqs · checked 2026-11-16 |
| Word timestamps | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Voice design | Yes | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Emotion control | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| On-premises deployment | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Priority support | Yes | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Training opt-out | Yessourceyou may opt out of our use of your Content for training at any time by navigating to the 'Data use' menuelevenlabs.io/terms-of-use · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Organizations and teams | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Access until period end after cancelling | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Long-form synthesis | YessourceMost stable on long-form generationselevenlabs.io/docs/overview/models.md · checked 2026-11-16 | not stated on the pages we read | YessourceCloud TTS API, which supports streaming and long-form text.cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-16 | not stated on the pages we read |
| Pause control | not stated on the pages we read | not stated on the pages we read | YessourcePause control is available across all locales.cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-16 | not stated on the pages we read |
| Language parameter | YessourceLanguage code (ISO 639-1) used to enforce a language for the model and text normalization.elevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Shared credit pool | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Data deleted with the account | YessourceAll data is deleted from our systems permanently when you delete your account.elevenlabs.io/docs/eleven-api/resources/zero-rete… · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read | YessourceFollowing closure of your AWS account, we will delete Your Content in accordance with the technical documentation applicable to the Services.aws.amazon.com/service-terms · checked 2026-11-16 |
| Quota increase on request | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Nonverbal tags | Yes | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Volume discount | Yes | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Dedicated support channel | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Several contexts on one WebSocket | YessourceWSS Multi-Context WebSocketelevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Browser access tokens | YessourceThese tokens are valid for a limited time and can be used to connect to the API without exposing your API key, for example from the client side.elevenlabs.io/docs/api-reference/authentication · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| GDPR compliant | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| SSML | Yessourceenable_ssml_parsing boolean Optional Defaults to falseelevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-16 | not stated on the pages we read | NosourceChirp 3: HD voices doesn't support SSML input, speaking rate and pitch-audio parameters, and the A-Law audio encoding.cloud.google.com/text-to-speech/docs/list-voices-… · checked 2026-11-16 | YessourceSynthesizes UTF-8 input, plain text or SSML, to a stream of bytes.docs.aws.amazon.com/polly/latest/APIReference/API… · checked 2026-11-16 |
| Credits kept when you change plan | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | YessourceIf you upgrade to a paid plan, any remaining Free Tier credit balance will automatically apply to your AWS bills.aws.amazon.com/polly/pricing · checked 2026-11-16 |
| Encryption in transit and at rest | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | Yes |
| ISO 27001 | not stated on the pages we read | not stated on the pages we read | Yessourcefor standards such as ISO 27001, ISO 27017, ISO 27018, SOC 2, SOC 3, and PCI DSS.docs.cloud.google.com/document-ai/docs/security · checked 2026-11-16 | not stated on the pages we read |
| PCI DSS | not stated on the pages we read | not stated on the pages we read | Yessourcefor standards such as ISO 27001, ISO 27017, ISO 27018, SOC 2, SOC 3, and PCI DSS.docs.cloud.google.com/document-ai/docs/security · checked 2026-11-16 | not stated on the pages we read |
| Usage dashboard | Yessourcenavigating to the analytics tab within the Developers dashboardelevenlabs.io/pricing/api · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Automatic recharge | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Mixed languages in one sentence | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Voice localization | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| IP indemnity | not stated on the pages we read | not stated on the pages we read | YessourceGoogle will defend Customer and its Affiliates using the Services under Customer’s Account and indemnify themcloud.google.com/terms · checked 2026-11-16 | not stated on the pages we read |
| CCPA compliant | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
Voice quality (measured by us) 4 rows
| Feature | ElevenLabs Creator | OpenAI tts-1-hd | Google Cloud Chirp 3: HD voices | Amazon Web Services Neural voices |
|---|---|---|---|---|
| Korean accuracy (character error rate, lower is better) | not measured yet | not measured yet | not measured yet | not measured yet |
| English accuracy (word error rate, lower is better) | not measured yet | not measured yet | not measured yet | not measured yet |
| Time to first sound | not measured yet | not measured yet | not measured yet | not measured yet |
| Human-preference quality (how pleasant it sounds) | not measured | not measured | not measured | not measured |
Billing (stated on the official pages) 45 rows
Terms and ownership (stated on the official pages) 34 rows
Data and privacy (stated on the official pages) 33 rows
Limits (stated on the official pages) 23 rows
Voices and models (stated on the official pages) 34 rows
| Feature | ElevenLabs Creator | OpenAI tts-1-hd | Google Cloud Chirp 3: HD voices | Amazon Web Services Neural voices |
|---|---|---|---|---|
| Supported languages | count: 90; model: eleven_v4 | 57: Afrikaans, Arabic, Armenian, Azerbaijani, Belarusian, Bosnian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, … +45 more (the TTS model generally follows Whisper's language support; voices are optimized for English)sourceAfrikaans, Arabic, Armenian, Azerbaijani, Belarusian, Bosnian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Galician, German, Greek, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Italian, Japanese, Kannada, Kazakh, Korean, Latvian, Lithuanian, …developers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | count: 30 (Instant Custom Voice languages (counted from the table))sourceInstant Custom Voice is supported in the following languages:cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | 41: Arabic (arb), Arabic (Gulf) (ar-AE), Catalan (ca-ES), Chinese (Cantonese) (yue-CN), Chinese (Mandarin) (cmn-CN), Czech (cs-CZ), Danish (da-DK), Dutch (Belgian) (nl-BE), Dutch (nl-NL), English (Australian) (en-AU), English (British) (en-GB), English (Indian) (en-IN), … +29 more (languages listed…sourceThe following languages are supported by Amazon Polly and can be used to synthesize speech.docs.aws.amazon.com/polly/latest/dg/supported-lan… · checked 2026-11-09 |
| Bidirectional text streaming | supported: Yes (WebSocket generates audio from partial text input)sourceThe Text-to-Speech WebSockets API is designed to generate audio from partial text inputelevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-09 | not stated on the pages we read | supported: Yes (streaming synthesis; streaming is only compatible with Chirp 3: HD voices)sourceStreaming Cloud Text-to-Speech is only compatible with Chirp 3: HD voices .cloud.google.com/text-to-speech/docs/create-audio… · checked 2026-11-09 | supported: Yes; operation: StartSpeechSynthesisStream; transport: HTTP/2 bidirectional (text sent while audio flows back, without either side waiting for the other to finish)sourceAmazon Polly provides a StartSpeechSynthesisStream operation that establishesdocs.aws.amazon.com/polly/latest/dg/bidirectional… · checked 2026-11-09 |
| Output formats | not stated on the pages we read | mp3, opus, aac, flac, wav, pcmsource- **MP3**: The default response format for general use cases. - **Opus**: For internet streaming and communication, low latency. - **AAC**: For digital audio compression, preferred by YouTube, Android, iOS. - **FLAC**: For lossless audio compression, favored by audio enthusiasts for archiving. - **…developers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | LINEAR16 (default), ALAW, MULAW, OGG_OPUS, PCM (Instant Custom Voice, streaming and batch)sourcestreaming : LINEAR16 (default), ALAW , MULAW , OGG_OPUS , PCMcloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | mp3, vorbis, pcm (audio stream formats named in the FAQ: MP3, Vorbis, raw PCM)sourceAmazon Polly supports MP3, Vorbis, and raw PCM audio stream formats.aws.amazon.com/polly/faqs · checked 2026-11-09 |
| Sdk languages | python, javascriptsourceREST API and SDKs for Python, JavaScript, and more.elevenlabs.io/pricing/api · checked 2026-11-09 | javascript, python, go, java, csharp, ruby (official client libraries; Speech examples also show curl and the openai CLI)sourceInstall an official client library from the [libraries page]developers.openai.com/api/reference/overview.md · checked 2026-11-09 | not stated on the pages we read | Java 2.x, JavaScript v3, .NET v4, C++, Go v2, Kotlin, PHP v3, Ruby v3, Rust, Swift (AWS SDKs that support the bidirectional streaming API)sourceAWS SDK for Java 2.x, JavaScript v3, .NET v4, C++, Go v2, Kotlin,docs.aws.amazon.com/polly/latest/dg/bidirectional… · checked 2026-11-09 |
| Tts models | eleven_v4, eleven_v4_turbo, eleven_v3, eleven_v3_conversational, eleven_multilingual_v2, eleven_flash_v2_5, eleven_flash_v2 | tts-1, tts-1-hd, gpt-4o-mini-ttssourceThe `tts-1`, `tts-1-hd`, and `gpt-4o-mini-tts` models are deprecated and willdevelopers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | gemini-3.1-flash-tts-preview, gemini-2.5-flash-tts, gemini-2.5-pro-tts, gemini-2.5-flash-lite-preview-ttssourceGemini-TTS includes the following available models:cloud.google.com/text-to-speech/docs/gemini-tts · checked 2026-11-09 | standard, neural, long-form, generativesourceValid Values: standard | neural | long-form | generativedocs.aws.amazon.com/polly/latest/APIReference/API… · checked 2026-11-09 |
| Clone sample seconds | not stated on the pages we read | max: 60 (audio samples for voices created from recordings)sourceThe audio samples must be 60 seconds or less.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | max: 10 (Instant Custom Voice: up to 10 seconds of single-channel reference audio)sourceRecord reference audio : Use your computer microphone to record up to 10cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | not stated on the pages we read |
| Custom voices max | not stated on the pages we read | 20 (voices created from recordings, per organization)sourceAt most 20 voices from recordings can be created per organization.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | unlimited (no limit on the number of voice cloning keys (Instant Custom Voice))sourceThere is no limit to the number of voice cloning keys that you can create,cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | not stated on the pages we read |
| Emotion control | not stated on the pages we read | Yes (gpt-4o-mini-tts, through prompts: accent, emotional range, intonation, impressions, speed of speech, tone, whispering)source`gpt-4o-mini-tts` lets you use prompts to control these aspects of speech:developers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | Yes (Gemini-TTS style control: natural-language prompts steer accent, tone and expressions incl. whisper, pace)sourceStyle control : Using natural language prompts, you can adapt thecloud.google.com/text-to-speech/docs/gemini-tts · checked 2026-11-09 | not stated on the pages we read |
| Voice cloning api | instant and professional voice clones can be created through the API | YessourceVoices from recordings must be created through an API request.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read |
| Voice design | Yes | Yes (prompt-to-voice is generally available to all API customers; prompt-generated voices work with GPT-Live only)sourcePrompt-to-voice is generally available to all API customers.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | Yes (Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS: voice design and voice replication in Preview, through the Agent Platform API and Agent Studio (not through the Cloud Text-to-Speech API))sourceGemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS are Generally Available (GA) through the Agent Platform API and Agent Studio (with voice design and voice replication in Preview). They aren't available through the Cloud Text-to-Speech API.cloud.google.com/text-to-speech/docs/gemini-tts · checked 2026-11-16 | not stated on the pages we read |
| Clone sample formats | not stated on the pages we read | mpeg, wav, ogg, aac, flac, webm, mp4sourceThe audio samples must be one of the following types: `mpeg`, `wav`, `ogg`, `aac`, `flac`, `webm`, or `mp4`.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | LINEAR16, PCM, MP3, M4A (Instant Custom Voice supported encoding formats)sourceLINEAR16 , PCM , MP3 , M4Acloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | not stated on the pages we read |
| Cross lingual synthesis | Multilingual v2 keeps voice quality and personality across all supported languages, maintaining the speaker's characteristics and accentsourceThe model delivers consistent voice quality and personality across all supported languages while maintaining the speaker's unique characteristics and accent.elevenlabs.io/docs/overview/models.md · checked 2026-11-09 | not stated on the pages we read | en-US voice cloning keys can synthesize de-DE, es-US, es-ES, fr-CA, fr-FR, pt-BR (Instant Custom Voice language transfer)sourceLanguage transfers : Voice cloning keys with locale en-US can synthesize output in the following locales: de-DE , es-US , es-ES , fr-CA , fr-FR , pt-BRcloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | some alphabets are not read by all voices (e.g. Cyrillic by English voices) unless phoneme mapping is usedsourceSome alphabets might not be available with all the voices (for example, Cyrillic might not be read at all by English voices) unless phoneme mapping is used.docs.aws.amazon.com/polly/latest/APIReference/API… · checked 2026-11-09 |
| Builtin voices | not stated on the pages we read | 9: alloy, ash, coral, echo, fable, onyx, nova, sage, shimmersourceThe `tts-1` and `tts-1-hd` models support a smaller set: `alloy`, `ash`, `coral`, `echo`, `fable`, `onyx`, `nova`, `sage`, and `shimmer`.developers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | count: 30 (Chirp 3: HD voices in 30 distinct styles across many languages)sourceThese voices are in 30 distinct styles across many languagescloud.google.com/text-to-speech/docs/list-voices-… · checked 2026-11-09 | 106: Aditi, Amy, Astrid, Bianca, Brian, Camila, Carla, Carmen, Celine, Chantal, Conchita, Cristiano, … +94 more (VoiceId valid values of SynthesizeSpeech (Aditi ... Tiffany))sourceValid Values: Aditi | Amy | Astriddocs.aws.amazon.com/polly/latest/APIReference/API… · checked 2026-11-09 |
| Cloned voice supported apis | TTS API: any library voice by ID, including professional, cloned and designed voicessourceYou can reference any voice in your library by ID, including professional voices, cloned voices, and voices you've designed.elevenlabs.io/pricing/api · checked 2026-11-09 | GPT-Live, Text to Speech API, Realtime API, Chat Completions API with audio outputsourceThese voices can be used with GPT-Live, the [Text to Speech API]developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read |
| First response latency ms | ms: 100; stat: median latency (excl. application & network); model: eleven_v4_turbosourceUltra-low latency (median latency of ~100ms†)elevenlabs.io/docs/overview/models.md · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Language override param | YessourceLanguage code (ISO 639-1) used to enforce a language for the model and text normalization.elevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Model api availability | Eleven v4: via the Text to Dialogue APIsourceEleven v4 is available via the [Text to Dialogue API]elevenlabs.io/docs/overview/models.md · checked 2026-11-09 | gpt-realtime-2.1-mini is not a supported model for /v1/audio/speech; migration is a change of API workflow to the Realtime APIsource`gpt-realtime-2.1-mini` isn't a supported model for `/v1/audio/speech`.developers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | Gemini 3.8 Flash TTS and Flash-Lite TTS (incl. voice design and voice replication): Preview through the Gemini Enterprise API only; not available through the Cloud Text-to-Speech APIsourceThey aren't available through the Cloud Text-to-Speech API.cloud.google.com/text-to-speech/docs/gemini-tts · checked 2026-11-09 | not stated on the pages we read |
| Model generation status | Eleven v3 (and v3 Conversational): previous generationsourceEleven v3 is our previous generation speech synthesis modelelevenlabs.io/docs/overview/models.md · checked 2026-11-09 | not stated on the pages we read | latest TTS models (LLM-powered) | not stated on the pages we read |
| Recommended replacement | eleven_flash_v2_5 (for eleven_turbo_v2_5)sourceFirst generation low-latency model (outclassed by Flash models)elevenlabs.io/docs/overview/models.md · checked 2026-11-09 | gpt-realtime-2.1-mini (Realtime API)source| **Jan 6, 2027** | `gpt-4o-mini-tts`, `gpt-4o-mini-tts-2025-03-20`, `gpt-4o-mini-tts-2025-12-15` | Deprecated | `gpt-realtime-2.1-mini` | Oct 1, 2026 |developers.openai.com/api/docs/deprecations.md · checked 2026-11-16 | not stated on the pages we read | not stated on the pages we read |
| Streaming synthesis start trigger | not stated on the pages we read | not stated on the pages we read | only after the client sends half-close (no more input) (Gemini-TTS streaming on Cloud Text-to-Speech API)sourcesynthesizing when the client sends Half-Close as a signal that it won't sendcloud.google.com/text-to-speech/docs/gemini-tts · checked 2026-11-09 | as input text arrives: text is sent as it becomes available (e.g. a model generating token by token) and audio is returned as it is synthesizedsourceYou send text as it becomes available, and Amazon Polly returns audio as it synthesizes,docs.aws.amazon.com/polly/latest/dg/bidirectional… · checked 2026-11-09 |
| Target use case | Eleven v4: character voiceovers, emotional dialogue, audiobook production, multilingual projects; content creation (professional content, audiobooks, video narration)sourcePerfect for long-form narration with complex emotional delivery.elevenlabs.io/docs/overview/models.md · checked 2026-11-09 | high quality text-to-speechsourceThe tts-1-hd model is optimized for high quality text-to-speech use cases.developers.openai.com/api/docs/models/tts-1-hd.md · checked 2026-11-09 | intended for Conversational Agents; examples: chat experiences, virtual assistants, customer service chatbots, interactive education, sales and pitches, storytime | not stated on the pages we read |
| Optimized for | eleven_flash_v2_5 / eleven_flash_v2: real-time use (~75 ms)sourceUltra-fast model optimized for real-time useelevenlabs.io/docs/overview/models.md · checked 2026-11-09 | qualitysourceText-to-speech model optimized for qualitydevelopers.openai.com/api/docs/models/tts-1-hd.md · checked 2026-11-09 | Gemini 3.1 Flash TTS and Gemini 2.5 Flash TTS: low latency, controllable, single- and multi-speaker audio for cost-efficient everyday applicationssourceLow latency, controllable, single- and multi-speaker Cloud TTS audio generation for cost-efficient everyday applicationscloud.google.com/text-to-speech/docs/gemini-tts · checked 2026-11-09 | not stated on the pages we read |
| Relative latency quality | Multilingual v2: higher latency and cost per character than Flash models, superior qualitysourceWhile it has a higher latency & cost per character than Flash models, it delivers superior quality for projects where lifelike speech is important.elevenlabs.io/docs/overview/models.md · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Ssml parsing | Yessourceenable_ssml_parsing boolean Optional Defaults to falseelevenlabs.io/docs/api-reference/text-to-speech/v… · checked 2026-11-09 | not stated on the pages we read | No (Chirp 3: HD voices do not support SSML input, speaking rate and pitch audio parameters, or the A-Law audio encoding)sourceChirp 3: HD voices doesn't support SSML input, speaking rate and pitch-audio parameters, and the A-Law audio encoding.cloud.google.com/text-to-speech/docs/list-voices-… · checked 2026-11-09 | Yes (input is UTF-8 plain text or SSML; SSML input must be valid, well-formed SSML)sourceSynthesizes UTF-8 input, plain text or SSML, to a stream of bytes.docs.aws.amazon.com/polly/latest/APIReference/API… · checked 2026-11-09 |
| Streaming model support | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read | generative engine only (StartSpeechSynthesisStream (bidirectional streaming) requires the generative engine and an AWS SDK that supports HTTP/2 event streams)sourceThis operation requires the generative engine and an AWS SDK that supports HTTP/2docs.aws.amazon.com/polly/latest/dg/bidirectional… · checked 2026-11-09 |
| Voice cloning eligibility | not stated on the pages we read | eligible customers only; contact sales; once enabled for the organization the Voices tab under Audio is availablesourceCustom voices created from recordings are limited to eligible customers.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | voice cloning (Instant Custom Voice) restricted to allow-listed users; request access from the sales teamsourceImportant: Access to Instant Custom Voice is restricted to allow-listed users.cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | not stated on the pages we read |
| Custom voice accents | not stated on the pages we read | English accents (gpt-live-1 custom voices; accent set in session.instructions)source`gpt-live-1` supports custom voices with English accents.developers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read |
| Web playground | not stated on the pages we read | OpenAI.fm interactive demo for trying the voices and the latest text-to-speech modelsourceHear and play with these voices in [OpenAI.fm](https://openai.fm), our interactive demodevelopers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read |
| Api backward compatibility policy | not stated on the pages we read | not stated on the pages we read | at least 12 months’ notice before significantly modifying a Customer-facing Google API in a backwards-incompatible manner (not for pre-GA offerings)sourceCustomer-facing Google API in a backwards-incompatible manner.cloud.google.com/terms · checked 2026-11-09 | not stated on the pages we read |
| Child voices allowed | no: voice recordings of anyone under 18 may not be used to create or train a voice modelsourceYou may not submit voice recordings of anyone under 18 to create or train a voice model.elevenlabs.io/use-policy · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read | not stated on the pages we read |
| Dated model snapshots | not stated on the pages we read | Yes (gpt-4o-mini-tts: default snapshot gpt-4o-mini-tts-2025-12-15; also gpt-4o-mini-tts-2025-03-20)sourceDefault snapshot: `gpt-4o-mini-tts-2025-12-15`developers.openai.com/api/docs/models/gpt-4o-mini… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read |
| Language quality tier | not stated on the pages we read | voices are currently optimized for EnglishsourceVoices are currently optimized for English.developers.openai.com/api/docs/guides/text-to-spe… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read |
| Voice cloning rpm | not stated on the pages we read | not stated on the pages we read | 10 (voice cloning key creation per minute per project (GenerateVoiceCloningKeyRequestsPerMinutePerProject))sourceYou can create 10 voice cloning keys per minute per project.cloud.google.com/text-to-speech/docs/chirp3-insta… · checked 2026-11-09 | not stated on the pages we read |
| Voice design candidates | not stated on the pages we read | 1sourceIt doesn't include audio or a previewdevelopers.openai.com/api/docs/guides/custom-voic… · checked 2026-11-09 | not stated on the pages we read | not stated on the pages we read |
Support and service (stated on the official pages) 4 rows
Other features (stated on the official pages) 31 rows
A transcript error can come from the transcriber and not from the voice. The same transcriber is used for every plan, so the order between plans is comparable and the absolute value is not. How pleasant a voice sounds is not measured.
Accuracy is measured by transcribing each plan's audio with one local speech-to-text model and comparing the text with the sentence; a number written another way (digits for words) counts as an error. How pleasant a voice sounds is not measured.