Voice and speech models

Duration and resolution, side by side

Only the models whose vendor publishes a figure appear here. "Not established" means no page we could retrieve stated one.

All 29

• Aura-2 Deepgram

  Deepgram's current TTS model, with 37 English voices plus voices in Spanish, Dutch, German, Italian, French and Japanese - seven languages total. It is priced per 1,000 characters and positi

• Azure AI Speech text-to-speech (Neural, Neural HD, MAI voices) Microsoft (Azure AI Speech)

  Microsoft's enterprise speech synthesis service, with 200+ prebuilt neural voices across 80+ locales plus tiers for HD voices, multilingual voices, multi-talker voices and MAI-Voice models w

• Chatterbox Multilingual Resemble AI

  An open-weights, 500M-parameter zero-shot TTS model from Resemble AI covering 23+ languages, released under the MIT licence. It clones a voice from a short reference clip without retraining

• Chirp 3: HD voices (Google Cloud Text-to-Speech) Google

  Google Cloud's current-generation HD voice set, with 28 voices available across 50+ languages and locales. It supports pace control from 0.25x to 2x, pause markup tags, custom pronunciations

• Deepdub (eTTS, Speech-to-Speech, Live) Deepdub

  Enterprise AI voice and dubbing stack covering expressive text-to-speech with emotional control, speech-to-speech voice conversion and translation, voice cloning that preserves speaker ident

• Eleven Flash v2.5 ElevenLabs

  The low-latency ElevenLabs model, documented at roughly 75ms model latency with a 40,000-character request limit. It covers 32 languages and is priced at half the rate of the higher-quality

• Eleven Multilingual v2 ElevenLabs

  The older ElevenLabs workhorse model, described in the docs as the most stable choice for long-form generation. It covers 29 languages and accepts up to 10,000 characters per request. Still

• Eleven v3 ElevenLabs

  ElevenLabs' most expressive text-to-speech model, aimed at emotionally varied narration rather than lowest latency. It supports natural multi-speaker dialogue and a large language set. Chara

• ElevenLabs Agents ElevenLabs

  A hosted conversational voice stack that chains a fine-tuned speech-to-text model, a selectable LLM, low-latency TTS and a proprietary turn-taking model. It includes workflow building, RAG k

• ElevenLabs Dubbing (v2) ElevenLabs

  End-to-end dubbing that transcribes, translates and regenerates speech in the source speaker's voice across 90+ languages, with dialect variants in v2. It detects multiple speakers including

• ElevenLabs Voice Changer (Multilingual STS v2) ElevenLabs

  Speech-to-speech conversion that keeps the performance of a recorded take - timing, emphasis, delivery - and replaces the timbre with a different voice. Two models are documented: a 29-langu

• ElevenLabs Voice Cloning (Instant and Professional) ElevenLabs

  Two cloning paths: Instant Voice Cloning estimates a voice from 1-2 minutes of audio with no verification, and Professional Voice Cloning fine-tunes a dedicated model from 30-180 minutes of

• Fish Audio S2.1 Pro Fish Audio

  Fish Audio's recommended production TTS model, covering 83 languages with natural-language delivery control using bracket syntax and multi-speaker dialogue. Pricing is a flat $15 per 1M UTF-

• Gemini native TTS Google

  Speech generation built into Gemini models, driven by natural-language style instructions rather than SSML. It offers 30 voice options across 60+ auto-detected languages, and a multi-speaker

• HeyGen Video Translation HeyGen

  Translates an existing video into another language, optionally regenerating the speaker's mouth movements to match the new audio. Offered in a faster Speed mode (audio-only or lip-sync) and

• Kokoro-82M hexgrad (Kokoro)

  A very small open-weights TTS model - 82M parameters, StyleTTS2 architecture with an ISTFTNet vocoder - released under Apache-2.0. Version 1.0 ships 54 voices across 8 language codes at 24kH

• LMNT TTS LMNT

  A low-latency TTS API with voice cloning from a 5-second recording and 31 languages including mid-sentence language switching. Streaming latency is stated as 150-200ms. Pricing is per 1,000

• MiniMax Speech 2.8 (HD and Turbo) MiniMax (Hailuo AI)

  MiniMax's current expressive TTS line, announced January 2026, in an HD variant for quality and a Turbo variant for speed. It supports roughly 40 languages via a language_boost parameter wit

• Murf Gen2 (and Falcon 2) Murf AI

  Murf offers two API models: Gen2 for studio-quality customisable synthesis and Falcon 2, a streaming model documented at sub-100ms latency for conversational use. The library covers 150+ voi

• Octave 2 Hume AI

  An LLM-based TTS model that interprets the meaning of the text to choose pronunciation, pitch and tempo rather than reading flatly. It supports designing a voice from a text description, clo

• OpenAI transcription (gpt-4o-transcribe, gpt-transcribe, Whisper) OpenAI

  A family of speech-to-text models spanning batch and realtime use: Whisper and gpt-4o-transcribe for files, gpt-4o-transcribe-diarize for speaker-separated transcripts, and gpt-live-transcri

• RWS AI Dubbing and Voiceover (formerly Papercup) RWS (Papercup)

  AI dubbing and voiceover service delivered as part of RWS's localisation offering. Papercup, previously an independent AI dubbing startup, no longer operates as a standalone product: papercu

• Rask AI Rask AI

  Video localisation platform that translates video into 135 languages, clones voices in 32 of them, and optionally lip-syncs the speaker to the new audio. Higher tiers add multi-speaker lip-s

• Respeecher Voice Marketplace (speech-to-speech and TTS) Respeecher

  Speech-to-speech voice conversion that maps one performer's recorded delivery onto a different voice while keeping emotion and dynamics, plus a TTS path. Voices are licensed through a market

• Scribe v2 ElevenLabs

  Speech-to-text model covering 90+ languages with word- and character-level timestamps and speaker diarization for up to 32 speakers. It adds entity detection and redaction, audio event taggi

• Sonic 3.5 Cartesia

  Cartesia's current TTS model, documented at sub-90ms first-byte latency across 42 languages. It is tuned for conversational delivery and for reading alphanumerics - order numbers, phone numb

• XTTS-v2 Coqui (defunct)

  An open-weights multilingual TTS and voice cloning model covering 17 languages at 24kHz, still widely used for local voice cloning. Coqui Inc. shut down in January 2024, so the model and the

• gpt-4o-mini-tts OpenAI

  OpenAI's current text-to-speech model, steerable with natural-language instructions about how a line should be delivered. It replaces the older tts-1 and tts-1-hd models, which are still lis

• gpt-realtime-2.1 OpenAI

  OpenAI's speech-to-speech model for live conversation, taking audio in and producing audio out with tool use. A mini variant trades quality for roughly a third of the cost. A separate gpt-re

Other categories

• Video generation 62
• Image generation 20
• Music and sound 20
• Editing and post 11
• Upscaling and restoration 8
• Lip sync and avatars 8

Everything in the directory