Voice & Audio

Text-to-speech

Tool id	Vendor	Credentials	Status	Divinci usage
browser-local Free text-to-speech using your browser's built-in Web Speech API. Uses system voices, works offline.	Browser	Divinci-managed	available	—
cartesia-sonic Premium text-to-speech with 40ms latency. Features emotion control, laughter tags, and instant voice cloning.	Cartesia	Divinci-managed	available	—
cloudflare-aura-1 High-quality context-aware text-to-speech powered by Deepgram on Cloudflare Workers AI. 12 voices, multiple audio formats.	Cloudflare	Divinci-managed	available	—
cloudflare-aura-2 Enhanced quality text-to-speech with improved pronunciation and natural prosody. 40 English and 10 Spanish voices.	Cloudflare	Divinci-managed	available	—
cloudflare-melotts Multi-lingual text-to-speech by MyShell AI. Supports 16+ languages with natural-sounding voices.	Cloudflare	Divinci-managed	available	—
deepgram-tts Deepgram Aura TTS with 100+ voices across 7 languages, low latency, and pronunciation control.	Deepgram	BYOK	available	—
elevenlabs-tts Highly expressive, natural ElevenLabs voices with 29+ languages and instant voice cloning support.	ElevenLabs	BYOK	available	—
openai-tts OpenAI speech (gpt-4o-mini-tts) with steerable delivery and 10 expressive built-in voices.	OpenAI	BYOK	available	—
vertex-ai-chirp3 Google's highest-fidelity Chirp 3 HD voices. Eight expressive voices available across 30+ locales.	Google	Divinci-managed	available	—
vertex-ai-neural2 High-quality neural text-to-speech voices from Google Cloud. Natural-sounding with 1M free characters per month.	Google	Divinci-managed	available	—
vertex-ai-standard Basic text-to-speech voices from Google Cloud. Affordable option with 4M free characters per month.	Google	Divinci-managed	available	—
vertex-ai-studio Premium studio-grade text-to-speech voices from Google Cloud. Professional quality for production audio.	Google	Divinci-managed	available	—

Usage figures for this category are not collected yet — the weekly job populates them.

cloudflare-aura-2 is the default. browser-local uses the listener's own device — no synthesis bill and no audio leaving the browser, at the cost of whatever voices that device happens to have.

`@cf/deepgram/aura-2-en` is a Cloudflare call, not a Deepgram one

The default voice is Deepgram's model hosted on Cloudflare Workers AI. It consumes no Deepgram credits and needs no Deepgram key, despite the deepgram in the model string. deepgram-tts is the separate tool that really does call api.deepgram.com. Switching the default to it changes whose bill the audio lands on and nothing a listener can hear.

Transcription

Tool id	Vendor	Credentials	Status	Divinci usage
@cf/openai/whisper-large-v3-turbo Opensource speech to text	OpenAI	Divinci-managed	available	—
@openai/whisper-1 Speech to text straight from the source	OpenAI	BYOK	available	—
deepgram-nova-3 Deepgram's Nova-3 speech-to-text — fast, accurate, multilingual	Deepgram	BYOK	available	—

Usage figures for this category are not collected yet — the weekly job populates them.

Whisper on Workers AI is the default; Deepgram Nova 3 and OpenAI's hosted Whisper are BYOK alternatives. Speech-to-text has a fallback router, so a provider failing mid-job falls through rather than losing the job.

Speaker diarization

Tool id	Vendor	Credentials	Status	Divinci usage
@divinci-ai/pyannote-segmentation Identifies speaker during time durations	Pyannote	Divinci-managed	available	—
@google-cloud/speech Speech to text including speakers	Google	Divinci-managed	available	—
@pyannote/pyannote-segmentation Identifies speaker during time durations	Pyannote	BYOK	available	—

Usage figures for this category are not collected yet — the weekly job populates them.

Diarization answers who spoke when, and runs alongside transcription rather than instead of it. Pyannote is available both on Divinci's infrastructure and on your own key.

Voice cloning & voiceprints

Tool id	Vendor	Credentials	Status	Divinci usage
@cartesia/voice-clone Creates an instant voice clone from a short audio clip (minimum 5 seconds)	Cartesia	BYOK	available	—
@google/chirp3-instant-custom-voice Creates a Chirp 3 voice cloning key from ~10s of reference audio plus a voice-talent consent recording. Requires Google allow-list access.	Google	Divinci-managed	available	—

Usage figures for this category are not collected yet — the weekly job populates them.

Tool id	Vendor	Credentials	Status	Divinci usage
@pyannote/voiceprint Creates a voice embedding from an audio sample for speaker identification	Pyannote	BYOK	available	—

Usage figures for this category are not collected yet — the weekly job populates them.

A voice clone synthesises new speech in a given voice. A voiceprint identifies a speaker across recordings. They are different capabilities with very different consent implications.

Both need the speaker's permission, not just their audio

Cloning a voice from a recording you have is technically easy and, in a growing number of jurisdictions, unlawful without explicit consent. Voiceprints are biometric identifiers and attract biometric-privacy law on their own terms. Neither is a place to move fast.

See also

• Voice, Phone & SMS
• Recall.ai connector — meeting capture with transcripts.