🎙️ ASR — Kabyle

Backends: Sunbird Whisper (Kabyle-optimized, GPU, timestamps) · Dolphin ONNX (Kabyle-optimized, CPU, greedy, no timestamps)

ASR Backend

Sample from audio/

Upload or record audio

Max duration (minutes)

1 30

Language: Forced Kabyle | Hardware: ZeroGPU

Transcription

SRT Subtitles (time-aligned)

Detected Language / Info

Segments

Download files

Tip: Add .wav / .mp3 / .flac / .ogg / .m4a files to the audio/ folder in this Space repo and restart to see them as pickable samples.

Backends at a glance

Sunbird Whisper — Kabyle-optimized, GPU, chunked long-form with timestamps (SRT/VTT).
Dolphin ONNX — Kabyle (kab-NULL), CPU, greedy decoding, plain text only — good for a quick side-by-side check without waiting on GPU.