Model Overview
This collection contains large size versions of Conformer-CTC (around 120M parameters) trained on NeMo German ASRSet with over 2000 hours of German speech. The model transcribes speech in lower case German alphabet along with spaces and without punctuation. This model was trained with the English model as initialization.
Model Architecture
Conformer-CTC model is a non-autoregressive variant of Conformer model [1] for Automatic Speech Recognition which uses CTC loss/decoding instead of Transducer. You may find more info on the detail of this model here: Conformer-CTC Model.
Training
The NeMo toolkit [3] was used for training the models for over several hundred epochs. These model are trained with this example script and this base config.
The tokenizers for these models were built using the text transcripts of the train set with this script.
You may find more info on how to train and use language models for ASR models here: ASR Language Modeling
Datasets
All the models in this collection are trained on a composite dataset (NeMo ASRSET) comprising of over two thousand hours of cleaned German speech:
- MCV7.0 567 hours
- MLS 1524 hours
- VoxPopuli 214 hours
Performance
The performance of Automatic Speech Recognition models is measuring using Word Error Rate. Since this dataset is trained on multiple domains and a much larger corpus, it will generally perform better at transcribing audio in general.
The latest model obtains the following greedy scores on the following evaluation datasets -
| Version | Decoding | MCV7.0 dev | MCV7.0 Test | Comment |
|---|---|---|---|---|
| 1.5.0 | Greedy | 5.84 | 6.68 | |
| 1.5.0_lm | 4-gram LM | 5.23 | 6.03 | beam_width=256, n_gram_alpha=1, n_gram_beta=1 |
| Version | Decoding | MLS dev | MLS test | Comment |
|---|---|---|---|---|
| 1.5.0 | Greedy | 3.85 | 4.63 | |
| 1.5.0_lm | 4-gram LM | 3.53 | 4.23 | beam_width=256, n_gram_alpha=0.5, n_gram_beta=0.5 |
| Version | Decoding | Vox Populi dev | Vox Populi Test | Comment |
|---|---|---|---|---|
| 1.5.0 | Greedy | 12.56 | 10.51 | |
| 1.5.0_lm | 4-gram LM | 11.69 | 9.82 | beam_width=256, n_gram_alpha=0.5, n_gram_beta=0.5 |
Note that these evaluation datasets have been filtered and preprocessed to only contain German alphabet characters and are removed of punctuation.
How to Use this Model
The model is available for use in the NeMo toolkit [3], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.
Automatically load the model from NGC
import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.EncDecCTCModelBPE.from_pretrained(model_name="stt_de_conformer_ctc_large")
Transcribing text with this model
python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py \
pretrained_name="stt_de_conformer_ctc_large" \
audio_dir=""
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.
Limitations
Since this model was trained on publically available speech datasets, the performance of this model might degrade for speech which includes technical terms, or vernacular that the model has not been trained on. The model might also perform worse for accented speech.
References
[1] Conformer: Convolution-augmented Transformer for Speech Recognition
[2] Google Sentencepiece Tokenizer
Licence
License to use this model is covered by the NGC TERMS OF USE 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.