File size: 3,749 Bytes
b6f1a69
 
13a1724
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
64fc74a
13a1724
 
 
64fc74a
13a1724
 
b6f1a69
 
0e33a0c
 
 
 
 
13a1724
b6f1a69
13a1724
 
 
 
 
d49eee6
13a1724
b6f1a69
13a1724
b6f1a69
13a1724
 
b6f1a69
e924b6a
b6f1a69
63eedef
b6f1a69
13a1724
 
b6f1a69
13a1724
b6f1a69
13a1724
b6f1a69
13a1724
b6f1a69
13a1724
b6f1a69
13a1724
 
 
 
b6f1a69
13a1724
 
 
b6f1a69
13a1724
 
b6f1a69
13a1724
 
 
 
b6f1a69
13a1724
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
---
library_name: transformers
license: mit
language: fr
datasets:
- Cnam-LMSSC/vibravox
metrics:
- per
tags:
- audio
- automatic-speech-recognition
- speech
- phonemize
- phoneme
model-index:
- name: Wav2Vec2-base French finetuned for Speech-to-Phoneme by LMSSC
  results:
  - task:
      name: Speech-to-Phoneme
      type: automatic-speech-recognition
    dataset:
      name: Vibravox["temple_vibration_pickup"]
      type: Cnam-LMSSC/vibravox
      args: fr
    metrics:
    - name: Test PER, in-domain training |
      type: per
      value: 14.2
---

<p align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/65302a613ecbe51d6a6ddcec/zhB1fh-c0pjlj-Tr4Vpmr.png" style="object-fit:contain; width:280px; height:280px;" >
</p>


# Model Card 

- **Developed by:** [Cnam-LMSSC](https://huggingface.co/Cnam-LMSSC)
- **Model type:** [Wav2Vec2ForCTC](https://huggingface.co/transformers/v4.9.2/model_doc/wav2vec2.html#transformers.Wav2Vec2ForCTC)
- **Language:** French
- **License:** MIT
- **Finetuned from model:** [facebook/wav2vec2-base-fr-voxpopuli-v2](https://huggingface.co/facebook/wav2vec2-base-fr-voxpopuli-v2)
- **Finetuned dataset:** `temple_vibration_pickup` audio of the `speech_clean` subset of [Cnam-LMSSC/vibravox](https://huggingface.co/datasets/Cnam-LMSSC/vibravox) (see [VibraVox paper on arXiV](https://arxiv.org/abs/2407.11828))
- **Samplerate for usage:** 16kHz

## Output

As this model is specifically trained for a speech-to-phoneme task, the output is sequence of [IPA-encoded](https://en.wikipedia.org/wiki/International_Phonetic_Alphabet) words, without punctuation.
If you don't read the phonetic alphabet fluently, you can use this excellent [IPA reader website](http://ipa-reader.xyz) to convert the transcript back to audio synthetic speech in order to check the quality of the phonetic transcription.

## Link to phonemizer models trained on other body conducted sensors : 

An entry point to all **phonemizers** models trained on different sensor data from the [Vibravox dataset](https://huggingface.co/datasets/Cnam-LMSSC/vibravox) is available at [https://huggingface.co/Cnam-LMSSC/vibravox_phonemizers](https://huggingface.co/Cnam-LMSSC/vibravox_phonemizers).  

### Disclaimer
Each of these models has been trained for a **specific non-conventional speech sensor** and is intended to be used with **in-domain data**. The only exception is the headset microphone phonemizer, which can certainly be used for many applications using audio data captured by airborne microphones.

Please be advised that using these models outside their intended sensor data may result in suboptimal performance.

## Training procedure

The model has been finetuned for 10 epochs with a constant learning rate of *1e-5*. To reproduce experiment please visit [jhauret/vibravox](https://github.com/jhauret/vibravox).

## Inference script : 

```python
import torch, torchaudio
from transformers import AutoProcessor, AutoModelForCTC
from datasets import load_dataset

processor = AutoProcessor.from_pretrained("Cnam-LMSSC/phonemizer_temple_vibration_pickup")
model = AutoModelForCTC.from_pretrained("Cnam-LMSSC/phonemizer_temple_vibration_pickup")
test_dataset = load_dataset("Cnam-LMSSC/vibravox", "speech_clean", split="test", streaming=True)

audio_48kHz = torch.Tensor(next(iter(test_dataset))["audio.temple_vibration_pickup"]["array"])
audio_16kHz = torchaudio.functional.resample(audio_48kHz, orig_freq=48_000, new_freq=16_000)

inputs = processor(audio_16kHz, sampling_rate=16_000, return_tensors="pt")
logits = model(inputs.input_values).logits
predicted_ids = torch.argmax(logits,dim = -1)
transcription = processor.batch_decode(predicted_ids)

print("Phonetic transcription : ", transcription)
```