Ahmed Wasfy

Model · Speech recognition

Nawah-Parakeet-60M

Ahmed Wasfy

A 62.7M-parameter Parakeet-style Token-and-Duration Transducer for Arabic speech recognition — FastConformer encoder, TDT objective — trained from scratch with no pretrained encoder and no external language model. On the six test sets of the it reaches 31.34 average WER / 13.62 CER, which places it 8th of 38 evaluated systems — ahead of Voxtral-Small-24B, gemma-4-E4B, Qwen3-ASR-1.7B, nvidia-conformer-ctc-large-arabic and every Whisper model, at a fraction of their size. Every model above it is between 1B and 30B parameters, or a commercial API. Averages are recomputed over these six sets for every model so the ordering is consistent. Baseline figures are the leaderboard's published numbers.…

Dataset · Speech recognition

unnamed-split-ar

Ahmed Wasfy

The data/ar/ partition of espnet/yodas3, cut from long recordings (median ~5 minutes) into clips of 5-45 seconds, each with its own transcript. Ready to batch for ASR training without further segmentation. Audio is Opus 48 kbps mono at 16 kHz in Ogg. Upstream timestamps are YouTube caption cues, not utterance segments. Measured on a shard of data/ar: 71.8% of consecutive cues strictly overlap (median gap -2.36 s): start + duration is how long a cue stays on screen, not when its speech ends. Only 9.7% of cue boundaries have a real silence gap >= 0.3 s, so cutting on caption boundaries puts most cuts inside continuous speech. Snapping caption boundaries to the quietest nearby window does not…

Dataset · Speech recognition

unnamed-ar

Ahmed Wasfy

The data/ar/ partition of espnet/yodas3, repackaged as parquet with the audio inline so it loads without a script: Every row is one long-form recording (median ~5 minutes) with its full transcript. Audio container only. Upstream ships Opus in a WebM/Matroska container, which libsndfile cannot open. Here it is Opus in Ogg, remuxed with ffmpeg -c:a copy — the Opus packets are copied untouched, so the audio is bit-identical to upstream. 48 kHz. Metadata joined. Upstream keeps audio in data/ar/audio/.tar and the 12 metadata columns in data/ar/metadata/.parquet; this version carries both in one row. Dialect columns added (see below). Nothing was filtered, resampled, re-encoded, segmented, or…