anteju commited on
Commit
a6d54f6
1 Parent(s): 1a15dbf

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +117 -0
README.md CHANGED
@@ -1,3 +1,120 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  license: cc-by-4.0
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - hr
4
+ library_name: nemo
5
+ datasets:
6
+ - ParlaSpeech-HR v1.0
7
+ thumbnail: null
8
+ tags:
9
+ - automatic-speech-recognition
10
+ - speech
11
+ - audio
12
+ - Transducer
13
+ - Conformer
14
+ - Transformer
15
+ - pytorch
16
+ - NeMo
17
+ - hf-asr-leaderboard
18
  license: cc-by-4.0
19
  ---
20
+
21
+ # NVIDIA Conformer-Transducer Large (Croatian)
22
+
23
+ <style>
24
+ img {
25
+ display: inline;
26
+ }
27
+ </style>
28
+
29
+ | [![Model architecture](https://img.shields.io/badge/Model_Arch-Conformer--CTC-lightgrey#model-badge)](#model-architecture)
30
+ | [![Model size](https://img.shields.io/badge/Params-120M-lightgrey#model-badge)](#model-architecture)
31
+ | [![Language](https://img.shields.io/badge/Language-hr-lightgrey#model-badge)](#datasets) |
32
+
33
+
34
+ This model transcribes speech in lowercase Croatian alphabet including spaces, and is trained on around 1665 hours of Croatian speech data.
35
+ It is a "large" variant of Conformer-Transducer, with around 120 million parameters.
36
+ See the [model architecture](#model-architecture) section and [NeMo documentation](https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/models.html#conformer-transducer) for complete architecture details.
37
+ It is also compatible with NVIDIA Riva for [production-grade server deployments](#deployment-with-nvidia-riva).
38
+
39
+
40
+ ## Usage
41
+
42
+ 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.
43
+
44
+ To train, fine-tune or play with the model you will need to install [NVIDIA NeMo](https://github.com/NVIDIA/NeMo). We recommend you install it after you've installed latest PyTorch version.
45
+
46
+ ```
47
+ pip install nemo_toolkit['all']
48
+ ```
49
+
50
+ ### Automatically instantiate the model
51
+
52
+ ```python
53
+ import nemo.collections.asr as nemo_asr
54
+ asr_model = nemo_asr.models.EncDecRNNTBPEModel.from_pretrained("nvidia/stt_hr_conformer_transducer_large")
55
+ ```
56
+
57
+ ### Transcribing using Python
58
+ Simply do:
59
+ ```
60
+ asr_model.transcribe(['<your_audio>.wav'])
61
+ ```
62
+
63
+ ### Transcribing many audio files
64
+
65
+ ```shell
66
+ python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py
67
+ pretrained_name="nvidia/stt_hr_conformer_transducer_large"
68
+ audio_dir="<DIRECTORY CONTAINING AUDIO FILES>"
69
+ ```
70
+
71
+ ### Input
72
+
73
+ This model accepts 16 kHz single-channel audio as input.
74
+
75
+ ### Output
76
+
77
+ This model provides transcribed speech as a string for a given audio sample.
78
+
79
+ ## Model Architecture
80
+
81
+ Conformer-Transducer model is an autoregressive variant of Conformer model [1] for Automatic Speech Recognition which uses Transducer loss/decoding. You may find more info on the detail of this model here: [Conformer-Transducer Model](https://docs.nvidia.com/deeplearning/nemo/user-guide/docs/en/main/asr/models.html#conformer-transducer).
82
+
83
+ ## Training
84
+
85
+ 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/asr_ctc/speech_to_text_rnnt_bpe.py) and this [base config](https://github.com/NVIDIA/NeMo/blob/main/examples/asr/conf/conformer/conformer_transducer_bpe.yaml).
86
+
87
+ 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).
88
+
89
+ The vocabulary we use contains 27 characters:
90
+ ```python
91
+ ['a', 'b', 'c', 'č', 'ć', 'd', 'đ', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'r', 's', 'š', 't', 'u', 'v', 'z', 'ž']
92
+ ```
93
+
94
+ Full config can be found inside the .nemo files.
95
+
96
+ ### Datasets
97
+
98
+ All the models in this collection are trained on ParlaSpeech-HR v1.0 Croatian dataset, which contains around 1665 hours of training data after data cleaning, 2.2 hours of developement and 2.3 hours of test data.
99
+
100
+ ## Performance
101
+
102
+ 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.
103
+
104
+ | Version | Tokenizer | Vocabulary Size | Dev WER | Test WER | Train Dataset |
105
+ |---------|-----------------------|-----------------|---------|----------|---------------------|
106
+ | 1.11.0 | SentencePiece Unigram | 128 | X.YZ | X.YZ | ParlaSpeech-HR v1.0 |
107
+
108
+ You may use language models (LMs) and beam search to improve the accuracy of the models.
109
+
110
+ ## Limitations
111
+
112
+ Since the model is trained just on ParlaSpeech-HR v1.0 dataset, the performance of this model might degrade for speech which includes terms, or vernecular that the model has not been trained on. The model might also perform worse for accented speech.
113
+
114
+ ## References
115
+
116
+ - [1] [Conformer: Convolution-augmented Transformer for Speech Recognition](https://arxiv.org/abs/2005.08100)
117
+
118
+ - [2] [Google Sentencepiece Tokenizer](https://github.com/google/sentencepiece)
119
+
120
+ - [3] [NVIDIA NeMo Toolkit](https://github.com/NVIDIA/NeMo)