Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,160 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- zgh
|
4 |
+
- kab
|
5 |
+
- shi
|
6 |
+
- rif
|
7 |
+
- tzm
|
8 |
+
license: cc-by-4.0
|
9 |
+
library_name: nemo
|
10 |
+
datasets:
|
11 |
+
- mozilla-foundation/common_voice_17_0
|
12 |
+
thumbnail: null
|
13 |
+
tags:
|
14 |
+
- automatic-speech-recognition
|
15 |
+
- speech
|
16 |
+
- audio
|
17 |
+
- TDT
|
18 |
+
- FastConformer
|
19 |
+
- Transducer
|
20 |
+
- NeMo
|
21 |
+
- pytorch
|
22 |
+
model-index:
|
23 |
+
- name: stt_zgh_fastconformer_transducer_small
|
24 |
+
results:
|
25 |
+
- task:
|
26 |
+
type: Automatic Speech Recognition
|
27 |
+
name: automatic-speech-recognition
|
28 |
+
dataset:
|
29 |
+
name: Mozilla Common Voice 17.0
|
30 |
+
type: mozilla-foundation/common_voice_17_0
|
31 |
+
config: zgh
|
32 |
+
split: test
|
33 |
+
args:
|
34 |
+
language: zgh
|
35 |
+
metrics:
|
36 |
+
- name: Test WER
|
37 |
+
type: wer
|
38 |
+
value: 72.44
|
39 |
+
- task:
|
40 |
+
type: Automatic Speech Recognition
|
41 |
+
name: automatic-speech-recognition
|
42 |
+
dataset:
|
43 |
+
name: Mozilla Common Voice 17.0
|
44 |
+
type: mozilla-foundation/common_voice_17_0
|
45 |
+
config: zgh
|
46 |
+
split: test
|
47 |
+
args:
|
48 |
+
language: zgh
|
49 |
+
metrics:
|
50 |
+
- name: Test CER
|
51 |
+
type: cer
|
52 |
+
value: 26.56
|
53 |
+
- task:
|
54 |
+
type: Automatic Speech Recognition
|
55 |
+
name: automatic-speech-recognition
|
56 |
+
dataset:
|
57 |
+
name: Mozilla Common Voice 17.0
|
58 |
+
type: mozilla-foundation/common_voice_17_0
|
59 |
+
config: kab
|
60 |
+
split: test
|
61 |
+
args:
|
62 |
+
language: kab
|
63 |
+
metrics:
|
64 |
+
- name: Test WER
|
65 |
+
type: wer
|
66 |
+
value: 39.78
|
67 |
+
- task:
|
68 |
+
type: Automatic Speech Recognition
|
69 |
+
name: automatic-speech-recognition
|
70 |
+
dataset:
|
71 |
+
name: Mozilla Common Voice 17.0
|
72 |
+
type: mozilla-foundation/common_voice_17_0
|
73 |
+
config: kab
|
74 |
+
split: test
|
75 |
+
args:
|
76 |
+
language: kab
|
77 |
+
metrics:
|
78 |
+
- name: Test CER
|
79 |
+
type: cer
|
80 |
+
value: 15.81
|
81 |
+
metrics:
|
82 |
+
- wer
|
83 |
+
- cer
|
84 |
+
pipeline_tag: automatic-speech-recognition
|
85 |
+
---
|
86 |
+
## Model Overview
|
87 |
+
|
88 |
+
<DESCRIBE IN ONE LINE THE MODEL AND ITS USE>
|
89 |
+
|
90 |
+
## NVIDIA NeMo: Training
|
91 |
+
|
92 |
+
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.
|
93 |
+
```
|
94 |
+
pip install nemo_toolkit['asr']
|
95 |
+
```
|
96 |
+
|
97 |
+
## How to Use this Model
|
98 |
+
|
99 |
+
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.
|
100 |
+
|
101 |
+
### Automatically instantiate the model
|
102 |
+
|
103 |
+
```python
|
104 |
+
import nemo.collections.asr as nemo_asr
|
105 |
+
asr_model = nemo_asr.models.ASRModel.from_pretrained("ayymen/stt_zgh_fastconformer_transducer_small")
|
106 |
+
```
|
107 |
+
|
108 |
+
### Transcribing using Python
|
109 |
+
First, let's get a sample
|
110 |
+
```
|
111 |
+
wget https://dldata-public.s3.us-east-2.amazonaws.com/2086-149220-0033.wav
|
112 |
+
```
|
113 |
+
Then simply do:
|
114 |
+
```
|
115 |
+
asr_model.transcribe(['2086-149220-0033.wav'])
|
116 |
+
```
|
117 |
+
|
118 |
+
### Transcribing many audio files
|
119 |
+
|
120 |
+
```shell
|
121 |
+
python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py pretrained_name="ayymen/stt_zgh_fastconformer_transducer_small" audio_dir="<DIRECTORY CONTAINING AUDIO FILES>"
|
122 |
+
```
|
123 |
+
|
124 |
+
### Input
|
125 |
+
|
126 |
+
This model accepts 16000 KHz Mono-channel Audio (wav files) as input.
|
127 |
+
|
128 |
+
### Output
|
129 |
+
|
130 |
+
This model provides transcribed speech as a string for a given audio sample.
|
131 |
+
|
132 |
+
## Model Architecture
|
133 |
+
|
134 |
+
<ADD SOME INFORMATION ABOUT THE ARCHITECTURE>
|
135 |
+
|
136 |
+
## Training
|
137 |
+
|
138 |
+
The model was trained for 42 epochs on a NVIDIA GeForce RTX 4050 Laptop GPU.
|
139 |
+
|
140 |
+
### Datasets
|
141 |
+
|
142 |
+
Common Voice 17 *kab* and *zgh* splits plus bible readings in Tachelhit and Tarifit.
|
143 |
+
|
144 |
+
## Performance
|
145 |
+
|
146 |
+
Metrics are computed on the cleaned, non-punctuated test sets of *zgh* and *kab* (converted to Tifinagh).
|
147 |
+
|
148 |
+
## Limitations
|
149 |
+
|
150 |
+
<DECLARE ANY POTENTIAL LIMITATIONS OF THE MODEL>
|
151 |
+
|
152 |
+
Eg:
|
153 |
+
Since this model was trained on publicly 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.
|
154 |
+
|
155 |
+
|
156 |
+
## References
|
157 |
+
|
158 |
+
<ADD ANY REFERENCES HERE AS NEEDED>
|
159 |
+
|
160 |
+
[1] [NVIDIA NeMo Toolkit](https://github.com/NVIDIA/NeMo)
|