Vocalix speech platform · TTS + STT · v1

Speech that doesn’t drift under you.

Text to speech and speech to text for Hindi, Gujarati, Marathi, Tamil and Indian English. Pin a model version and it stays pinned. Test for free and get byte-identical audio on every run.

START FREE — $5 CREDIT READ DOCS

No card · test keys free forever · wallet untouched

Mel spectrogram · echo-1.0.0 · 22 050 Hzstreaming

5languages

14voices

<200msfirst byte

ap-south-1mumbai region

Neveraudio used for training

POST /v1/tts text → speechPOST /v1/stt speech → texthi · gu · mr · ta · en-IN five languagesecho-turbo $8 / 1M charsecho-1 $15 / 1M charsecho-pro $40 / 1M charsscribe-turbo $0.006 / minscribe-1 $0.012 / minvx_test_ free forever, deterministiccache hit billed at 20%402 before the job runs, not at invoice timeRFC 7807 stable error codesdefault_model_pin resolves once, at key creationOpenAI SDK change the base URLPOST /v1/tts text → speechPOST /v1/stt speech → texthi · gu · mr · ta · en-IN five languagesecho-turbo $8 / 1M charsecho-1 $15 / 1M charsecho-pro $40 / 1M charsscribe-turbo $0.006 / minscribe-1 $0.012 / minvx_test_ free forever, deterministiccache hit billed at 20%402 before the job runs, not at invoice timeRFC 7807 stable error codesdefault_model_pin resolves once, at key creationOpenAI SDK change the base URL

01 — Try it

Both directions, in the browser, before you sign up.

The same two endpoints your code will call. Nothing here is a video.

Response

endpointPOST /v1/tts

x-vocalix-model-versionecho-1.0.0

voice_idmeera-hi

first byte186 ms

characters39

charge$0.0000 · test

status200 OK

02 — Proof

Guarantees, not adjectives.

Everyone’s demo clip sounds good. These are the properties that decide whether it survives a release cycle and a finance review.

Determinism

Run it twice. Same bytes.

Test keys take the deterministic path — same text, voice, format, speed, pitch and seed in, byte-identical audio out. Zero GPU, zero cost, snapshot tests that stop flaking.

run 1 · seed 42

87e020bd84c82ec8047b8a77a59a2f62

200 · $0.0000

run 2 · seed 42

87e020bd84c82ec8047b8a77a59a2f62

✓ identical

run 3 · seed 43

86e01f2a85c8305b037b88e4a69a30f5

≠ one field changed

press it as often as you like

Code-switching

Hinglish stays Hinglish

आपका parcel deliver हो गया

Mid-sentence script changes are the default case here, not an edge case — in both directions.

Version pinning

Pinned is pinned

Set default_model_pinwhen you mint a key. It resolves once, at creation — a new release can’t change how your IVR sounds on a Friday.

Billing

Rejected before it runs

The wallet is reserved up front. Short on credit and you get 402 with the exact shortfall — never a surprise at invoice time.

Cache

Repeats cost a fifth

Identical requests are served from cache and billed at 20% of rate. Your greeting is synthesised once, not once per call.

Keys

Shown once, revoked instantly

Argon2id at rest, per-key rate limit and concurrency, effective on the next request. A leaked key can’t mint its own replacement.

03 — Spec

Everything the API will tell you, up front.

Languages	hi · gu · mr · ta · en-IN
Directions	POST /v1/tts · POST /v1/stt
Output formats	wav_<rate> · pcm_<rate> · mulaw_<rate>
Speed / pitch	0.25–4.0× · ±20 semitones
Transcript extras	word timestamps · diarisation · srt · vtt
Errors	RFC 7807 · stable codes
Compatibility	OpenAI audio.speech + audio.transcriptions

04 — API

Two endpoints. Nothing else to learn.

Send text, get audio. Send audio, get text. Same key, same wallet, same error contract. Already on the OpenAI SDK? Change the base URL and keep your code.

# text → speech
curl https://api.vocalix.ai/v1/tts \
  -H "Authorization: Bearer $VOCALIX_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text":"आपका parcel आज शाम तक deliver हो जाएगा।",
       "voice_id":"meera-hi","model":"echo-1","seed":42}' \
  --output reply.wav

# speech → text
curl https://api.vocalix.ai/v1/stt \
  -H "Authorization: Bearer $VOCALIX_KEY" \
  -F file=@support-call-4471.wav \
  -F model=scribe-1 -F language=auto -F diarize=true

05 — Price

Per character. Per second. No seats.

No monthly minimum and no credits that expire. Start on $5.

Model	Rate
Text to speech — per 1M characters
echo-turbo Lowest latency, voice agents	$8 /1M
echo-1 The default. Balanced.	$15 /1M
echo-pro Most expressive, narration	$40 /1M
Speech to text — per minute of audio
scribe-turbo Realtime, telephony	$0.006 /min
scribe-1 Diarisation, word timestamps	$0.012 /min
test keys Deterministic, zero GPU	Free

✓ billed per second, rounded up✓ cache hits at 20%✓ $5 on signup, no card✓ credit never expires

Ship a voice your users recognise.

Five minutes to your first call. $5 of credit, no card, and test keys that stay free while you build.

CREATE ACCOUNT SEE THE API