Downloads · 30 days
12
43% of all-time downloads
livewhisper-ml-indicwhisper
itsskofficial/livewhisper-ml-indicwhisper
livewhisper-ml-indicwhisper is a automatic speech recognition model from itsskofficial. Use it when you need speech turned into text. It is set up for ctranslate2. The card lists the license as mit.
AI4Bharat's IndicWhisper Malayalam model (Whisper-medium fine-tuned on the Vistaar training sets), converted to CTranslate2 float16 for use in LiveWhisper, a Windows voice-dictation app. LiveWhisper ships without PyTo…
All-time downloads
28
Public
Repo size
1.5 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.bin1.5 GB · 100%
At a glance
- Task
- Automatic Speech Recognition
- Library
- ctranslate2
- License
- mit
- Access
- Public
- Created
- Sep 23, 2026
- Updated
- Sep 23, 2026
- SHA
- f0f2e27e
Model card
From the Hugging Face model README
IndicWhisper Malayalam, converted for LiveWhisper
AI4Bharat's IndicWhisper Malayalam model (Whisper-medium fine-tuned on the Vistaar training sets), converted to CTranslate2 float16 for use in LiveWhisper, a Windows voice-dictation app. LiveWhisper ships without PyTorch, so it downloads its speech models already converted (ADR 0007).
Nothing was changed but the conversion.
Accuracy
tests/bench_asr.py on FLEURS, through LiveWhisper's own decoding path
(word error rate, lower is better):
| this model | whisper-large-v3-malayalam-ct2 | large-v3 | |
|---|---|---|---|
| FLEURS test, native script | 50.6% | 61.3% | 114.6% |
| FLEURS test, delivered romanized | 49.3% | 58.8% | 108.7% |
| FLEURS dev, native script | 52.1% |
"delivered" is what the user is given: the transcript after romanization, scored against every Latin spelling attested in Google's Dakshina.
Credit
Original model and training data: AI4Bharat Vistaar (Bhogale et al., "Vistaar: Diverse Benchmarks and Training Sets for Indian Language ASR"), MIT licence. Built on OpenAI's Whisper medium.
Licence: MIT, as the original.
- Task
- Automatic Speech Recognition
- Library
- ctranslate2
- License
- mit
- Languages
- ml
- Created
- Sep 23, 2026
- Updated
- Sep 23, 2026