Deepgram Nova-3 vs LR-ASD
Both in the video & clipping category. Side-by-side — pick the one that fits your stack tonight.
The speech-to-text that actually gets word-level timestamps right.
- rating
- 5★
- tested
- ✓ loya-tested
- cost
- paid
- install
- needs-wiring
- stars
- 0
- updated
- 4d ago
You only transcribe short voice notes — use free Whisper locally.
The 2025 state-of-the-art for 'which face is actually talking.' Fast, tiny, accurate.
- rating
- 4★
- tested
- —
- cost
- free
- install
- sidecar
- stars
- 109
- updated
- 1y ago
You're not building a pipeline yourself. This is a research model, not a product.
why it matters · Deepgram Nova-3
Nova-3 is the transcription engine behind every podcast clipper that ships. You upload audio, get back text with per-word timestamps, speaker labels, and punctuation — the three things you need to cut a clip on a clean sentence boundary instead of mid-word. Costs about 26 cents an hour of audio. Free \$200 credit when you sign up, which gets you through your first 700+ hours before you pay anything. Way more accurate than Whisper on real podcast audio.
why it matters · LR-ASD
LR-ASD is the newest open-source active speaker detection model (Springer IJCV 2025 paper). It tells your video pipeline which person in a multi-face frame is actually talking. Accuracy beats the older TalkNet approach and it's 23 times lighter — fast enough to run on every frame, not just samples. If you're building your own clipping or auto-crop pipeline and accuracy matters more than a pre-built library, this is the one to drop in. MIT, free, Python.