Skip to main contentTroubleshooting pre-recorded uploads
Go through this guide to learn about common bugs and issues that might occur when transcribing audio with Lightning STT.
Low-quality transcripts
Symptoms: transcripts miss words or contain heavy noise.
Fixes:
- Resample audio to at least 16 kHz and keep it mono.
- Prefer lossless codecs (WAV, FLAC) whenever you control the capture pipeline.
- Maintain 128 kbps or higher bitrate for compressed formats.
- Normalize and denoise audio before uploading (see best practices).
Large file sizes
Symptoms: uploads stall or take too long to complete.
Fixes:
- Use compressed formats such as MP3 or OGG if WAV is oversized.
- Downsample anything above 16 kHz; speech rarely needs more.
- Split recordings longer than 10 minutes into smaller chunks.
- Keep bitrates near 128 kbps unless fidelity is critical.
Symptoms: API responses return format/codec errors immediately.
Fixes:
- Convert the file with FFmpeg to a supported format (WAV, MP3, FLAC, OGG, WebM).
- Make sure the file extension matches the actual codec/container.
- Validate codec compatibility—Linear PCM is the safest fallback.
Still stuck?
- Re-run uploads with
?word_timestamps=true to confirm the request reaches the model.
- Capture the failing
request_id and share it with [email protected].
