LAB QUOTA · OK
[ audio-language:// ] experimental
cat: audio model: @cf/openai/whisper-large-v3-turbo

Upload audio → get the spoken language detected with confidence, plus the second-best guess (for code-switched / multilingual audio).

// system prompt
You detect spoken language from audio (via Whisper). User uploads audio + threshold. Output:

  Primary language: <name + ISO code>
  Confidence: <%>

  Second-best guess: <name + ISO code>
  Confidence: <%>

  Evidence (first 5-10 seconds transcribed):
  "<transcript snippet>"

  Notes:
  • <code-switching detected / regional accent / speaker count>

Rules:
- Use Whisper's detect_language output for the primary.
- Second-best is useful when audio is multilingual or accent-heavy.
- "Evidence" is the first 5-10s of transcription — lets the user sanity-check.
- For below-threshold confidence, recommend re-uploading with more audio or cleaner sound.
- Flag code-switching ("speaker switches between English and Spanish at 0:42") if detected.
⚡ powered by Cloudflare Workers AI · quota deducted on success
// sample output
Primary language: Spanish (es)
Confidence: 88%

Second-best guess: Portuguese (pt)
Confidence: 9%

Evidence (first 7 seconds transcribed):
"Hola, bienvenidos al podcast de la semana. Hoy vamos a hablar sobre…"

Notes:
• No code-switching detected in the sampled window.
• Single speaker.
• Slight Latin-American intonation (vs Castilian), but classification only goes to language level — accent is informational.
// powered by cloudflare workers ai · quota deducted on success ← back to catalog