|
--- |
|
model-index: |
|
- name: parthiv11/stt_hi_conformer_ctc_large_v2 |
|
results: |
|
- task: |
|
type: automatic-speech-recognition |
|
dataset: |
|
name: Librispeech (clean) |
|
type: librispeech_asr |
|
config: other |
|
split: test |
|
args: |
|
language: en |
|
metrics: |
|
- type: wer |
|
value: 8.1 |
|
name: WER |
|
language: |
|
- hi |
|
metrics: |
|
- wer |
|
library_name: nemo |
|
pipeline_tag: automatic-speech-recognition |
|
--- |
|
## Model Overview |
|
|
|
This collection contains large size versions of Conformer-CTC (around 120M parameters) trained on ULCA & Europal with around ~2900 hours. The model transcribes speech in Hindi characters along with spaces for Hinglish speech. |
|
|
|
## Model Architecture |
|
|
|
Conformer-CTC model is a non-autoregressive variant of Conformer model for Automatic Speech Recognition which uses CTC loss/decoding instead of Transducer. You may find more info on the detail of this model [here](https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/models.html). |
|
|
|
## Training |
|
|
|
The NeMo toolkit [3] was used for training the models for over several hundred epochs. These model are trained with [this example script](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/speech_to_text_bpe.py) and [this base config](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/conf/conformer/conformer_ctc_bpe.yaml). |
|
|
|
The tokenizers for these models were built using the text transcripts of the train set with [this script](https://github.com/NVIDIA/NeMo/blob/main/scripts/tokenizers/process_asr_text_tokenizer.py). |
|
|
|
The checkpoint of the language model used as the neural rescorer. |
|
|
|
### Datasets |
|
|
|
All the models in this collection are trained on Hindi labelled dataset (~2900 hrs): |
|
|
|
- ULCA Hindi Corpus |
|
- Europal Dataset |
|
|
|
## Performance |
|
|
|
The list of the available models in this collection is shown in the following table. Performances of the ASR models are reported in terms of Word Error Rate (WER%) with greedy decoding and 6-Gram KenLM trained on AI4Bharat Corpus and Europal. |
|
|
|
| Decoding | Version | Tokenizer | Vocabulary Size | MUCS 2021 Blind Test* | IITM 2020 Eval Set | IITM 2020 Dev Set | Common Voice 6 Test* | Common Voice 7 Test* | Common Voice 8 Test* | |
|
|-----------------|---------|---------------------|-----------------|------------------------|--------------------|-------------------|----------------------|----------------------|----------------------| |
|
| Greedy | 1.10.0 | SentencePiece Unigram | 128 | 9.37%/2.74% | 12.93%/5.60% | 12.63%/5.49% | 13.16%/4.5% | 13.5%/5.2% | 14.37%/5.95% | |
|
| 6-Gram KenLM** | 1.10.0 | SentencePiece Unigram | 128 | 11.79%/3.35% | 15.96%/6.39% | 15.49%/6.25% | 17.05%/5.43% | 17.77%/6.23% | 19.18%/7.1% | |
|
|
|
*- Normalized and without special characters and punctuation. |
|
|
|
**- KenLM with 128 beam size with n_gram_alpha=1.0, n_gram_beta=1.0. |
|
|
|
## How to Use this Model |
|
- Can also be used from NGC, intrution [here](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_hi_conformer_ctc_large). |
|
- Follow [colab](https://colab.research.google.com/drive/1mLWVCbe4JFnooDoQLG0_33Je0LXdCZjO?usp=sharing) to use it directly |
|
|
|
|
|
### Input |
|
This model accepts 16000 KHz Mono-channel Audio (wav files) as input. |
|
|
|
### Output |
|
This model provides transcribed speech as a string for a given audio sample. |
|
|
|
### Licence (Credit goes to Nvidia) |
|
License to use this model is covered by the [NGC TERMS OF USE](https://ngc.nvidia.com/legal/terms) unless another License/Terms Of Use/EULA is clearly specified. By downloading the public and release version of the model, you accept the terms and conditions of the [NGC TERMS OF USE](https://ngc.nvidia.com/legal/terms). |
