Speech to Text (STT)

Takes an audio file, returns the spoken content as text. 100% Whisper-compatible — if you're using OpenAI Whisper, you can switch to Qevron with a one-line base_url change.

Available models

Model	Use for	Note
solab-stt (recommended)	Mixed Turkish + English speech	Whisper-large-v3 derivative, faster-whisper
whisper-1	Via OpenAI	If an OpenAI-keyed channel exists

curl

curl https://app.qevron.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer sk-..." \
  -F "model=solab-stt" \
  -F "file=@/path/to/audio.wav" \
  -F "language=tr"

Python (openai SDK)

from openai import OpenAI

client = OpenAI(api_key="sk-...", base_url="https://app.qevron.ai/v1")

with open("recording.wav", "rb") as f:
    resp = client.audio.transcriptions.create(
        model="solab-stt",
        file=f,
        language="en",   # recommended — prevents drifting to the wrong language
    )
print(resp.text)

Language hint

Always pass language. Auto-detection can mis-classify on very short clips; a one-word file is enough to throw off auto-detect.

Supported formats

wav, mp3, m4a, ogg, flac, webm, mpga. Mono 16 kHz preferred; multi-channel is downmixed server-side.

Whisper-specific parameters

-F "prompt=Names that appear: Koray, Arpanet, Qevron"
-F "temperature=0"
-F "response_format=verbose_json"   # returns segments + timestamps

• prompt — initial context; specialised names / jargon here improve accuracy.
• temperature — default 0; raise toward 1 only in very specific scenarios.
• response_format — text (default), json, verbose_json (timestamped segments), srt, vtt.

Streaming

"stream": true returns delta chunks. Python openai>=1.50 supports it; not all SDKs do yet (see streaming notes).

stream = client.audio.transcriptions.create(
    model="solab-stt", file=open("rec.wav","rb"), stream=True
)
for delta in stream:
    print(delta.text, end="", flush=True)

Troubleshooting

• Empty response → check language is correct and the file format is supported.
• Wrong language → set language=en / language=tr explicitly.
• Too slow → split clips longer than 30 s; Whisper works in 30-second windows internally.

Next: Telephony (CalleKit) · TTS — text to speech