# Pascal

> One voice API. Read this index, then fetch the markdown URL for the page you need. To search from a coding agent, connect the read-only docs MCP. It cannot create a key or spend credits.

- Docs MCP: https://www.pascalai.co/.well-known/mcp
- Same server: https://www.pascalai.co/api/mcp
- Account MCP: https://api.pascalai.co/mcp
- API: https://api.pascalai.co
- Human docs: https://pascalai.co/api/docs

## Pages

- [Start here](https://pascalai.co/llm/start.md): Auth, live sample rates, file limits, and the rules an agent must not invent.
- [Routes](https://pascalai.co/llm/routes.md): The HTTP and WebSocket paths on api.pascalai.co.
- [Voice agents](https://pascalai.co/llm/voice-agents.md): Save a Deepgram profile, then open one socket. session.profile is required.
- [Whisper Large V3 Turbo](https://pascalai.co/llm/whisper-large-v3-turbo.md): Whisper Large V3 Turbo transcription.
- [Whisper Large V3](https://pascalai.co/llm/whisper-large-v3.md): Full Whisper Large V3, including translate to English.
- [Parakeet TDT 0.6B v3](https://pascalai.co/llm/parakeet-tdt-0.6b-v3.md): NVIDIA Parakeet batch ASR. 25 European languages.
- [Parakeet TDT 0.6B v2](https://pascalai.co/llm/parakeet-tdt-0.6b-v2.md): NVIDIA Parakeet English TDT.
- [Parakeet CTC 0.6B](https://pascalai.co/llm/parakeet-ctc-0.6b.md): NVIDIA Parakeet English CTC. Word timestamps from token times.
- [Parakeet RNNT 0.6B](https://pascalai.co/llm/parakeet-rnnt-0.6b.md): NVIDIA Parakeet English RNNT. Word timestamps from token times.
- [Canary 1B Flash](https://pascalai.co/llm/canary-1b-flash.md): NVIDIA Canary Flash. 25 European languages. Word timestamps.
- [Canary 1B v2](https://pascalai.co/llm/canary-1b-v2.md): NVIDIA Canary 1B v2. 25 European languages. Word timestamps.
- [Qwen3 ASR 1.7B](https://pascalai.co/llm/qwen3-asr-1.7b.md): Qwen3 batch ASR. 30 languages. Accepts a prompt.
- [Cohere Transcribe](https://pascalai.co/llm/cohere-transcribe-03-2026.md): Cohere batch transcription. Language is required.
- [Pascal Transcribe](https://pascalai.co/llm/pascal-transcribe.md): Speech-to-text. $0.18 per audio hour, or $0.20 per audio hour with speaker diarization, plus the $0.05 request fee.
- [Sarvam Saaras](https://pascalai.co/llm/sarvam-saaras.md): Saaras v4. $0.005 / min. Diarization is $0.0075 / min. language and language_code are the same field.
- [Sarvam Saaras Live](https://pascalai.co/llm/sarvam-saaras-live.md): Live Saaras. $0.005 / min. PCM16, μ-law, or A-law at 8000 or 16000 Hz.
- [Sarvam Bulbul](https://pascalai.co/llm/sarvam-bulbul.md): Bulbul v3. $30 per million characters.
- [Sarvam Bulbul Live](https://pascalai.co/llm/sarvam-bulbul-live.md): Live Bulbul. $30 per million characters. linear16 at 24 kHz.
- [Sarvam Voice Clone](https://pascalai.co/llm/sarvam-clone.md): Clone a voice from one clip. $30 per million characters.
- [Sarvam Dubbing](https://pascalai.co/llm/sarvam-dubbing.md): $0.40 per source minute for each target language. Billed per second.
- [Nemotron 3.5 ASR Streaming](https://pascalai.co/llm/nemotron-3.5-asr-streaming-0.6b.md): Live speech-to-text. WebSocket only. $0.00333 / min plus the $0.05 request fee.
- [Pascal Transcribe Live](https://pascalai.co/llm/pascal-transcribe-live.md): Live speech-to-text. WebSocket only. $0.22 per audio hour, or $0.25 per audio hour with speaker diarization, plus the $0.05 request fee.
- [Deepgram Nova-3](https://pascalai.co/llm/deepgram-nova-3.md): File speech-to-text at Deepgram Pay As You Go. $0.0043 per minute for one language, $0.0052 per minute when language is multi. File diarization is included.
- [Deepgram Nova-3 Streaming](https://pascalai.co/llm/deepgram-nova-3-streaming.md): Live Nova-3. $0.0048 per minute for one language, $0.0058 per minute when language is multi. Diarization adds $0.0020 per minute.
- [Deepgram Flux English](https://pascalai.co/llm/deepgram-flux-en.md): Live English Flux. $0.0065 per minute. Turn detection is part of the model.
- [Deepgram Flux Multilingual](https://pascalai.co/llm/deepgram-flux-multi.md): Live multilingual Flux. $0.0078 per minute.
- [Deepgram Aura-2](https://pascalai.co/llm/deepgram-aura-2.md): Aura-2 speech. $30 per million characters. Every published Aura-2 voice.
- [Deepgram Aura-1](https://pascalai.co/llm/deepgram-aura-1.md): Aura-1 speech. $15 per million characters. Every published Aura-1 voice.
- [Deepgram Flux](https://pascalai.co/llm/deepgram-flux.md): Flux speech. $45 per million characters. Every published Flux voice.
- [Deepgram Aura-2 Live](https://pascalai.co/llm/deepgram-aura-2-live.md): Live Aura-2. $30 per million characters. Same voices as the file model.
- [Deepgram Aura-1 Live](https://pascalai.co/llm/deepgram-aura-1-live.md): Live Aura-1. $15 per million characters.
- [Deepgram Flux Live](https://pascalai.co/llm/deepgram-flux-live.md): Live Flux speech. $45 per million characters.
- [Piper TTS Realtime](https://pascalai.co/llm/piper-en-us-lessac-medium.md): Live Piper Lessac Medium. WebSocket only. $0.00333 / min plus the $0.05 request fee.
- [Matcha TTS Realtime](https://pascalai.co/llm/matcha-en-ljspeech.md): Live Matcha LJSpeech. WebSocket only. $0.00333 / min plus the $0.05 request fee.
- [Pocket TTS Realtime](https://pascalai.co/llm/pocket-tts.md): Live Pocket TTS. WebSocket only. $0.00333 / min plus the $0.05 request fee.
- [Kokoro](https://pascalai.co/llm/kokoro.md): Text to speech. Voices across nine languages, plus speed.
- [Chatterbox Turbo](https://pascalai.co/llm/chatterbox-turbo.md): Chatterbox Turbo. Tags such as [laugh] go in text.
- [Chatterbox Original](https://pascalai.co/llm/chatterbox-original.md): Chatterbox original. exaggeration and cfg_weight.
- [Chatterbox Multilingual v3](https://pascalai.co/llm/chatterbox-multilingual-v3.md): Chatterbox multilingual. language_id, exaggeration, and cfg_weight.
- [Qwen3 TTS 0.6B](https://pascalai.co/llm/qwen3-tts-0.6b.md): Qwen3 CustomVoice 0.6B. language and voice.
- [Qwen3 TTS 1.7B](https://pascalai.co/llm/qwen3-tts-1.7b.md): Qwen3 CustomVoice 1.7B. language, voice, and instruct.
- [OmniVoice](https://pascalai.co/llm/omnivoice.md): OmniVoice. language, num_step, guidance_scale, and speed.
- [VoxCPM2](https://pascalai.co/llm/voxcpm2.md): VoxCPM2 text to speech. Native generate is 48 kHz mono.
- [Pocket TTS Clone](https://pascalai.co/llm/pocket-tts-clone.md): Clone from one reference clip, then speak new text.
- [Chatterbox Turbo Clone](https://pascalai.co/llm/chatterbox-turbo-clone.md): Chatterbox Turbo zero-shot clone.
- [Chatterbox Clone](https://pascalai.co/llm/chatterbox-clone.md): Chatterbox original zero-shot clone. Same exaggeration and CFG as speech.
- [Chatterbox Multilingual Clone](https://pascalai.co/llm/chatterbox-multilingual-clone.md): Chatterbox multilingual clone. language_id, exaggeration, and cfg_weight.
- [Qwen3 TTS 0.6B Clone](https://pascalai.co/llm/qwen3-tts-0.6b-clone.md): Qwen3 Base 0.6B clone. language plus optional x_vector_only.
- [Qwen3 TTS 1.7B Clone](https://pascalai.co/llm/qwen3-tts-1.7b-clone.md): Qwen3 Base 1.7B clone. language plus optional x_vector_only.
- [OmniVoice Clone](https://pascalai.co/llm/omnivoice-clone.md): OmniVoice clone. Same knobs as speech, plus optional instruct.
- [CosyVoice3 Clone](https://pascalai.co/llm/cosyvoice3-clone.md): Upload a short clip of the speaker, type what that clip says, then type the new words.
- [VoxCPM2 Clone](https://pascalai.co/llm/voxcpm2-clone.md): VoxCPM2 clone. reference_text is ignored.
- [AuK Flash Clone](https://pascalai.co/llm/auk-flash-clone.md): AuK Flash zero-shot clone. Send model auk-flash-clone.
- [AuK Base Clone](https://pascalai.co/llm/auk-base-clone.md): AuK Base zero-shot clone. Send model auk-base-clone.
- [Qwen3 TTS Voice Design](https://pascalai.co/llm/qwen3-tts-voice-design.md): Describe a voice, then speak text. No reference clip.
- [OmniVoice Voice Design](https://pascalai.co/llm/omnivoice-voice-design.md): OmniVoice. voice_description is tags only.
- [VoxCPM2 Voice Design](https://pascalai.co/llm/voxcpm2-voice-design.md): Free-form voice description, then speak text.
- [AuK Flash Voice Design](https://pascalai.co/llm/auk-flash-voice-design.md): Describe a voice, then speak text. No reference clip. Send model auk-flash-voice-design.
- [AuK Base Voice Design](https://pascalai.co/llm/auk-base-voice-design.md): Describe a voice, then speak text. No reference clip. Send model auk-base-voice-design.
- [CosyVoice3 Voice Conversion](https://pascalai.co/llm/cosyvoice3-voice-conversion.md): Two clips. Source keeps the words. Reference is the new voice. No transcript and no new text.
- [VoxCPM2 Continuation](https://pascalai.co/llm/voxcpm2-continuation.md): Continue a spoken prompt in the same voice.
- [HTDemucs](https://pascalai.co/llm/htdemucs.md): Four-stem Demucs. overlap is the one extra runtime knob.
- [HTDemucs FT](https://pascalai.co/llm/htdemucs-ft.md): Fine-tuned HTDemucs. Same stems and overlap as htdemucs.
- [HTDemucs 6s](https://pascalai.co/llm/htdemucs-6s.md): Six-stem Demucs, including guitar and piano.
- [MDX Extra Q](https://pascalai.co/llm/mdx-extra-q.md): Demucs MDX Extra Q. Same overlap knob as HTDemucs.
- [MelBand RoFormer](https://pascalai.co/llm/melband-roformer.md): The stems list picks two-stem or four-stem.
- [BS-RoFormer 6-stem](https://pascalai.co/llm/bs-roformer-6stem.md): Six-stem RoFormer. No overlap field.
- [RNNoise](https://pascalai.co/llm/rnnoise.md): Speech denoise. Output options only.
- [DeepFilterNet 2](https://pascalai.co/llm/deepfilternet2.md): DeepFilterNet 2. attenuation_limit_db, post_filter, and LSNR scale.
- [DeepFilterNet 3](https://pascalai.co/llm/deepfilternet3.md): DeepFilterNet 3. Same extra knobs as DeepFilterNet 2.
- [ClearerVoice Studio](https://pascalai.co/llm/clearervoice-studio.md): MossFormer2 denoise. No DeepFilterNet knobs.
- [Resemble Denoise](https://pascalai.co/llm/resemble-denoise.md): Resemble denoise. Output options only.
- [VoiceFixer](https://pascalai.co/llm/voicefixer.md): Reconstructs damaged speech.