AI2026
cohere-transcribe
High-throughput Arabic/English transcription, batched properly.
- Live
Project factsبيانات المشروع
In short
A Python package for offline batch transcription with Cohere's 2B Arabic/English ASR model: optimised voice activity detection, multi-file GPU batching under bounded memory, subtitle output and optional word-level timestamps.
The compartments
- Role
- Contributor
- Year
- 2026
- Status
- Live
Built with
- Python
- Silero VAD
- GPU batching
- SRT / VTT
- Hugging Face
Where to find it
What it isما هو
Transcribing one file is a solved problem. Transcribing a directory of hundreds without running out of GPU memory, and without spending most of the run on silence, is not.
The package handles individual files, multiple paths and nested directories with bounded-memory batching, and can return results directly or publish plain text, approximate segment-timed subtitles, or word-timed subtitles.
تفريغ ملف واحد مسألة محلولة. أما تفريغ مجلد فيه مئات الملفات دون نفاد ذاكرة المعالج الرسومي، ودون إنفاق معظم زمن التشغيل على الصمت، فليس كذلك.
تعالج الحزمة الملفات المفردة والمسارات المتعددة والمجلدات المتشعبة بمعالجة دفعية محدودة الذاكرة، وتستطيع إعادة النتائج مباشرة أو إخراج نص عادي أو ترجمات بتوقيت تقريبي للمقاطع أو ترجمات بتوقيت على مستوى الكلمة.
Results
- Directory-scale batch transcription with bounded memory and subtitle output.
- تفريغ دفعي على مستوى المجلدات بذاكرة محدودة وإخراج ترجمات.
Challenges
Each one opens twice: what went wrong on the left, what I did on the right.
01
Most of a recording is silence
Problem
Feeding raw audio to the model spends a large share of the compute transcribing pauses, which is the main reason long-file throughput is bad.
What I did
Voice activity detection segments speech first, so only speech reaches the model. The batching is then built around those segments rather than around file boundaries.
Keep goingتابع
Next project
T3 CodeControl the coding agents on your machine, from anywhere.Open it →افتحه ←Shelf
All projectsAll 40, in six compartments.