File size: 1,527 Bytes
1ad58b4
f725a8a
 
0052610
 
 
 
d2f0b77
 
0052610
 
 
 
d2f0b77
0052610
 
 
 
 
 
 
d2f0b77
0052610
 
 
 
 
d2f0b77
0052610
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
# 🎙️ ASR Workspace

Place ASR baselines, training configs, and evaluation scripts here.

## ✅ Verified Pashto-Relevant ASR Models

### 🧠 OpenAI Whisper Large v3
- Model: [huggingface.co/openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3)
- Pashto validation: [OpenAI Whisper tokenizer map includes `"ps": "pashto"`](https://raw.githubusercontent.com/openai/whisper/main/whisper/tokenizer.py).
- Use in this repo: strong baseline and pseudo-labeling engine for bootstrapping.
- Applications: transcription, subtitle generation, dataset pre-labeling.

### 🌐 Meta MMS Coverage (ASR + TTS language support)
- Coverage page: [MMS language coverage](https://dl.fbaipublicfiles.com/mms/misc/language_coverage_mms.html)
- Pashto validation: row includes `pus` with ASR and TTS support.
- Use in this repo: multilingual transfer baseline when Pashto data is limited.
- Applications: low-resource ASR transfer experiments.

## ⚙️ Verified Inference Tooling

### 🚀 Faster-Whisper
- Repo: [github.com/SYSTRAN/faster-whisper](https://github.com/SYSTRAN/faster-whisper)
- Why useful: optimized Whisper inference for faster experimentation.
- Use in this repo: local transcription pipelines and benchmark generation speedups.

## 🧩 Integration Hints
- Keep all model/eval runs reproducible with command logs and commit hashes.
- Store evaluation outputs under [benchmarks/](../benchmarks/README.md) with model/version labels.
- Track WER/CER with dataset split and normalization policy references.