This collection contains large size versions of Conformer-CTC (around 120M parameters) trained on MCV-9.0 Catalan dataset.
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: Conformer-CTC Model.
The tokenizers  for these models were built using the text transcripts of the train set with this script, and can be found inside the .nemo files.
The vocabulary we use contains 44 characters:
[' ', "'", '-', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', '·', 'à', 'á', 'ç', 'è', 'é', 'í', 'ï', 'ñ', 'ò', 'ó', 'ú', 'ü', 'ı', '–', '—']
Full config can be found inside the .nemo files.
All the models in this collection are trained on MCV-9.0 Catalan dataset, which contains around 1203 hours training, 28 hours of development and 27 hours of testing speech audios.
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.
|Version||Tokenizer||Vocabulary Size||Dev WER||Test WER||Train Dataset|
|1.11.0||SentencePiece Unigram||128||4.70||4.27||MCV-9.0 Train set|
You may use language models (LMs) and beam search to improve the accuracy of the models, as reported in the follwoing table.
|Language Model||Test WER||Test WER w/ Oracle LM||Train Dataset||Settings|
|N-gram LM||3.77||1.54||MCV-9.0 Train set||N=6, beam_width=128, ngram_alpha=1.5, ngram_beta=2.0|
The model is available for use in the NeMo toolkit , and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.
import nemo.collections.asr as nemo_asr asr_model = nemo_asr.models.EncDecCTCModelBPE.from_pretrained(model_name="stt_ca_conformer_ctc_large")
python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py \ pretrained_name="stt_ca_conformer_ctc_large" \ audio_dir="[PATH_TO_AUDIO_FOLDER]"
This model accepts 16 kHz mono-channel audio (wav files) as input.
This model provides transcribed speech as a string for a given audio sample.
Since all models are trained on just MCV-9.0 dataset, 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.