Update model card

README.md

@@ -13,8 +13,8 @@ tags:
 
 # MME_STT_Model
 
-   Hindi speech to Roman-script Hinglish transcription. This model was fine-tuned with LoRA and the adapter
-   was merged into the weights, so it loads as a normal Transformers model with no PEFT.
+Hindi speech to Roman-script Hinglish transcription (lowercase, no punctuation). The model was fine-tuned
+with LoRA and the adapter was merged into the weights, so it loads as a normal Transformers model with no PEFT.
 
 ## Usage
 
@@ -25,23 +25,18 @@ from transformers import WhisperForConditionalGeneration, WhisperProcessor
 model = WhisperForConditionalGeneration.from_pretrained("MmeOG/MME_STT_Model", torch_dtype=torch.float16).to("cuda")
 processor = WhisperProcessor.from_pretrained("MmeOG/MME_STT_Model")
 
-audio, _ = librosa.load("sample.wav", sr=16000)
+audio, _ = librosa.load("sample.wav", sr=16000)   # 16 kHz mono, up to 30 seconds
 feats = processor(audio, sampling_rate=16000, return_tensors="pt").input_features.to("cuda", torch.float16)
 ids = model.generate(feats, language="hi", task="transcribe")
 print(processor.batch_decode(ids, skip_special_tokens=True)[0])
 ```
 
-## Training
-
-LoRA fine-tune on about 190k utterances: IndicVoices (about 92%), MUCS and UJS (about 4% each).
-Two epochs, batch size 16, learning rate 3e-5 (cosine), one A100 80GB.
-
 ## Evaluation
 
-Phonetic WER (pWER, lower is better) on the public set `RinggAI/ASR-Benchmarking-Dataset`
-(sampled 200 clips per dataset, 1,100 clips in total). The merged model reproduces the adapter's published scores to within 0.05 points.
+Phonetic WER (pWER, lower is better) on six public Hindi sets (1,100 clips). It is a lenient metric used
+to compare systems on the same clips, not an absolute accuracy figure.
 
-| Dataset | This model | Before fine-tuning) |
+| Dataset | This model | Before fine-tuning |
 |---|---|---|
 | fleurs | 18.45 | 20.15 |
 | commonvoice | 26.42 | 29.16 |
@@ -51,13 +46,11 @@ Phonetic WER (pWER, lower is better) on the public set `RinggAI/ASR-Benchmarking
 | mucs | 19.65 | 20.84 |
 | **all** | **19.64** | **21.21** |
 
-pWER is a lenient metric that compares text in a shared phonetic form, so it ranks systems on the
-same clips but should not be read as an absolute accuracy figure.
-
 ## Limitations
 
-- Not a streaming model: it transcribes after the speaker finishes.
-- Short utterances (under 2 seconds) have high error rates.
-- Accuracy drops sharply in loud background noise, especially babble.
-- Not evaluated on real phone-call recordings.
-- Commercial systems scored better on the same public benchmark.
+- Not validated on real phone calls.
+- Not a streaming model; it transcribes after the speaker finishes.
+- Accuracy drops in noise: about 5 to 7 points at 10 dB, and strong noise (0 dB) breaks it.
+- Occasional repetition loops (about 0.5% of clips); a repetition guard at decoding time is recommended.
+- Very short clips (under 2 seconds) have high error rates.
+- Outputs have no punctuation or capitalisation, and there is no speaker diarisation.