File size: 3,890 Bytes
97f6728
 
 
 
 
 
0d6ed0f
97f6728
 
0d6ed0f
7addc0e
0d6ed0f
97f6728
0d6ed0f
a5df75d
 
0d6ed0f
97f6728
0d6ed0f
97f6728
0d6ed0f
97f6728
0d6ed0f
97f6728
0d6ed0f
97f6728
0d6ed0f
 
97f6728
0d6ed0f
97f6728
0d6ed0f
 
e8704e5
97f6728
 
 
 
 
 
e8704e5
97f6728
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
90a6d44
 
 
 
 
 
 
 
e8704e5
90a6d44
 
 
 
 
 
 
e8704e5
e80389d
90a6d44
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
217f583
90a6d44
 
 
 
1acf81a
 
90a6d44
 
217f583
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
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
---
language: en
datasets:
- librispeech_asr
tags:
- speech

license: apache-2.0
---

# Wav2Vec2-Large-960h

[Facebook's Wav2Vec2](https://ai.facebook.com/blog/wav2vec-20-learning-the-structure-of-speech-from-raw-audio/)

The large model pretrained and fine-tuned on 960 hours of Librispeech on 16kHz sampled speech audio. When using the model
make sure that your speech input is also sampled at 16Khz.

[Paper](https://arxiv.org/abs/2006.11477)

Authors: Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, Michael Auli

**Abstract**

We show for the first time that learning powerful representations from speech audio alone followed by fine-tuning on transcribed speech can outperform the best semi-supervised methods while being conceptually simpler. wav2vec 2.0 masks the speech input in the latent space and solves a contrastive task defined over a quantization of the latent representations which are jointly learned. Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/other test sets. When lowering the amount of labeled data to one hour, wav2vec 2.0 outperforms the previous state of the art on the 100 hour subset while using 100 times less labeled data. Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 4.8/8.2 WER. This demonstrates the feasibility of speech recognition with limited amounts of labeled data.

The original model can be found under https://github.com/pytorch/fairseq/tree/master/examples/wav2vec#wav2vec-20.


# Usage

To transcribe audio files the model can be used as a standalone acoustic model as follows:

```python
 from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
 from datasets import load_dataset
 import soundfile as sf
 import torch
 
 # load model and tokenizer
 tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-large-960h")
 model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h")
 
 # define function to read in sound file
 def map_to_array(batch):
     speech, _ = sf.read(batch["file"])
     batch["speech"] = speech
     return batch
     
 # load dummy dataset and read soundfiles
 ds = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation")
 ds = ds.map(map_to_array)
 
 # tokenize
 input_values = tokenizer(ds["speech"][:2], return_tensors="pt", padding="longest").input_values  # Batch size 1
 
 # retrieve logits
 logits = model(input_values).logits
 
 # take argmax and decode
 predicted_ids = torch.argmax(logits, dim=-1)
 transcription = tokenizer.batch_decode(predicted_ids)
 ```
 
## Evaluation
 
This code snippet shows how to evaluate **facebook/wav2vec2-large-960h** on LibriSpeech's "clean" and "other" test data.
 
```python
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Tokenizer
import soundfile as sf
import torch
from jiwer import wer


librispeech_eval = load_dataset("librispeech_asr", "clean", split="test")

model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h").to("cuda")
tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-large-960h")

def map_to_array(batch):
    speech, _ = sf.read(batch["file"])
    batch["speech"] = speech
    return batch

librispeech_eval = librispeech_eval.map(map_to_array)

def map_to_pred(batch):
    input_values = tokenizer(batch["speech"], return_tensors="pt", padding="longest").input_values
    with torch.no_grad():
        logits = model(input_values.to("cuda")).logits

    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = tokenizer.batch_decode(predicted_ids)
    batch["transcription"] = transcription
    return batch

result = librispeech_eval.map(map_to_pred, batched=True, batch_size=1, remove_columns=["speech"])

print("WER:", wer(result["text"], result["transcription"]))
```

*Result (WER)*:

| "clean" | "other" |
|---|---|
| 2.8 | 6.3 |