Audio

The audio endpoints offer four operations: text-to-speech (TTS), voice management, speech-to-text (STT) and speech-to-English translation.

Text-to-speech (TTS)

POST/v1/audio/speech

Converts text into a spoken audio file.

Request parameters (JSON)

Field	Type	Required	Description
model	string	✓	TTS model (blab-fast, blab-pro, or one of their aliases)
input	string	✓	Text to speak
voice	string		A voice id (not a language code). Omit it for the engine default — see the Voice Roster
response_format	string		wav (default), mp3, pcm
speed	number		Speed between 0.25–4.0
quality	string		blab-fast only: standard / high / max
instruction	string		blab-pro only: natural-language style direction (e.g. "cheerful, slightly faster")
stream_format	string		For streamed output ("sse"); auto-bridged from stream: true

voice is not validated

An unrecognised voice id raises no error — the request returns 200 with the engine's default voice. Verify ids with the listing endpoint below. opus / aac / flac are not supported by these engines: blab-fast silently falls back to WAV, blab-pro returns 400.

Example

curl

curl https://app.qevron.ai/v1/audio/speech \
  -H "Authorization: Bearer $QEVRON_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "blab-fast",
    "input": "Hello, this audio was generated with Qevron.",
    "voice": "emma"
  }' --output output.wav

Python

resp = client.audio.speech.create(
    model="blab-fast",
    input="Hello, this audio was generated with Qevron.",
    voice="emma",
)
resp.stream_to_file("output.wav")

The response is binary audio data (Content-Type: audio/wav by default). When streaming, audio chunks arrive as SSE speech.audio.delta events.

Listing voices

GET/v1/audio/voices

Returns the voices a TTS model can use: the engine's built-in roster plus your own clones. Other tenants' clones are never listed.

Query parameters

Field	Type	Required	Description
model	string	✓	TTS model name (blab-fast, blab-pro, or an alias)

Example

curl -H "Authorization: Bearer $QEVRON_API_KEY" \
  "https://app.qevron.ai/v1/audio/voices?model=blab-fast"

Response

{
  "supports_custom": true,
  "voices": [
    {
      "id": "emma",
      "label": "Emma",
      "builtin": true,
      "lang": "en",
      "lang_key": "en",
      "label_en": "Emma",
      "gender": "F",
      "age": "adult",
      "category": "audiobooks",
      "categories": ["audiobooks"],
      "tags": ["warm", "narrative", "smooth", "calm", "mature"],
      "description": "Emma, uzun dinlemeler için sıcak ve sürükleyici bir anlatıcı.",
      "description_en": "Emma is a warm, immersive narrator for long-form listening."
    },
    { "id": "M1", "label": "Kerem — Genç", "builtin": true, "lang": null }
  ]
}

The two entries above are both verbatim from the live endpoint and show the two shapes you must handle: a fully catalogued roster voice, and an uncatalogued one (here blab-fast's generic default style M1 — note it carries no status, no lang_key and no catalogue fields).

Field dictionary

Field	Type	On every record?	Description
supports_custom	bool	✓ (root field)	Whether the engine can create custom (cloned) voices
id	string	✓	What you put in the voice field of /v1/audio/speech
label	string	✓	Display name (in its own language)
builtin	bool	✓	true = engine roster voice, false = tenant clone
status	string	no	"ready". Coverage is inconsistent — see the caveat below
lang	string	no	Language the voice was curated for (tr / en). Absent or null on uncatalogued records
lang_key	string	no	Same value and same coverage as lang
label_en	string	no	English display name (currently identical to label on every record)
gender	string	no	M (male) or F (female)
age	string	no	young / adult / mature
category	string	no	Use case it was curated for (e.g. news, audiobooks)
categories	string[]	no	Array form of category; currently always exactly one element
tags	string[]	no	Exactly five character tags (e.g. warm, crisp)
description	string	no	Short blurb (in its own language)
description_en	string	no	English blurb (best-effort; a few records still return Turkish text)

The full set of values, with every id: TTS Voice Roster.

Coverage caveat — catalogued vs uncatalogued records

Not every listed voice carries a catalogue card. Only the curated roster has gender / age / category / tags / description (blab-pro: 47 of 91 records; blab-fast: 40 of 60). The rest are tenant clones and blab-fast's generic M1–F5 default styles, which carry only id, label, builtin (plus status if the engine supplies it).

When you write code against this:

1. lang and lang_key have identical coverage — both are populated only on the catalogued roster. "lang_key is always populated" is not true; a language filter built on lang_keydrops exactly the same voices as one built on lang (i.e. it hides the user's own clones). The only difference is cosmetic: for uncatalogued blab-fast records lang is sent as null, while lang_key is not sent at all.
2. Your language filter must handle "unknown" — do not discard a voice because lang is missing.
3. Do not rely on status — blab-pro sends it on every record (91/91), blab-fast on only 10 of 60. Do not build a "ready" badge on it; if the field is absent, treat the voice as usable.

Errors

Case	Response
model not supplied	400 — {"error": {"message": "query param 'model' is required"}}
Model is not a local TTS model	404
Engine unreachable	502

Cloning a custom voice

POST/v1/audio/voices

Creates a custom voice for your group from a reference clip. Uses multipart/form-data. Instant cloning is available on blab-pro.

Field	In	Required	Description
model	query or form	✓	TTS model name
label	form	✓	Display name of the voice
file	form	✓	Reference audio clip (3–60 s)
transcript	form		The clip's exact text. Supplied, it upgrades blab-pro to prompt-continuation ("ultimate") cloning for sharper similarity; other engines ignore it

curl -X POST "https://app.qevron.ai/v1/audio/voices?model=blab-pro" \
  -H "Authorization: Bearer $QEVRON_API_KEY" \
  -F "label=Host Voice" \
  -F "file=@reference.wav"

The created voice's id is namespaced to your group (g<group>__<label>), so tenants can neither see nor collide with each other's voices. Use the returned id verbatim in the voice field — not the label you sent.

Deleting a voice

DELETE/v1/audio/voices/{id}

Deletes a custom voice you created. The model query parameter is required. You may only delete your own group's voices; built-ins and other tenants' voices return 403. Use it to reap ephemeral clones (for example the per-speaker clones created during dubbing).

curl -X DELETE "https://app.qevron.ai/v1/audio/voices/gacme__host_voice?model=blab-pro" \
  -H "Authorization: Bearer $QEVRON_API_KEY"

Speech-to-text (STT)

POST/v1/audio/transcriptions

Transcribes an audio file into text. Uses multipart/form-data.

Form fields

Field	Type	Required	Description
model	string	✓	STT model (e.g. solab-stt)
file	file	✓	Audio file (wav, mp3, m4a, ...)
language	string		Language code (e.g. tr, en)
response_format	string		json (default), verbose_json, text

Example

curl

curl https://app.qevron.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $QEVRON_API_KEY" \
  -F "model=solab-stt" \
  -F "file=@recording.wav" \
  -F "language=en"

Python

with open("recording.wav", "rb") as f:
    resp = client.audio.transcriptions.create(
        model="solab-stt",
        file=f,
        language="en",
    )
print(resp.text)

Response

{ "text": "Hello, this is a test recording." }

With verbose_json you additionally get language, duration and segments.

Speech-to-English translation

POST/v1/audio/translations

Uses the same form fields as transcriptions, but the output text is translated into English. The language field is not needed.