File size: 4,830 Bytes
9f87a17 b072afc 9f87a17 309d70f 9f87a17 91c912a 1856b7e 2e79ab5 bc05f0c 2e79ab5 d0a04de 2e79ab5 d0a04de 2e79ab5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 |
---
language:
- multilingual
- en
- de
- tr
- fa
- sv
- mn
- zh
- cy
- ca
- sl
- et
- id
- ar
- ta
- lv
- ja
datasets:
- common_voice
- multilingual_librispeech
- covost2
tags:
- speech
- xls_r
- automatic-speech-recognition
- xls_r_translation
pipeline_tag: automatic-speech-recognition
license: apache-2.0
widget:
- example_title: English
src: https://cdn-media.huggingface.co/speech_samples/common_voice_en_18301577.mp3
---
# Wav2Vec2-XLS-R-1B-EN-15
Facebook's Wav2Vec2 XLS-R fine-tuned for **Speech Translation.**
This is a [SpeechEncoderDecoderModel](https://huggingface.co/transformers/model_doc/speechencoderdecoder.html) model.
The encoder was warm-started from the [**`facebook/wav2vec2-xls-r-1b`**](https://huggingface.co/facebook/wav2vec2-xls-r-1b) checkpoint and
the decoder from the [**`facebook/mbart-large-50`**](https://huggingface.co/facebook/mbart-large-50) checkpoint.
Consequently, the encoder-decoder model was fine-tuned on 15 `en` -> `{lang}` translation pairs of the [Covost2 dataset](https://huggingface.co/datasets/covost2).
The model can translate from spoken `en` (Engish) to the following written languages `{lang}`:
`en` -> {`de`, `tr`, `fa`, `sv-SE`, `mn`, `zh-CN`, `cy`, `ca`, `sl`, `et`, `id`, `ar`, `ta`, `lv`, `ja`}
For more information, please refer to Section *5.1.1* of the [official XLS-R paper](https://arxiv.org/abs/2111.09296).
## Usage
### Demo
The model can be tested on [**this space**](https://huggingface.co/spaces/facebook/XLS-R-1B-EN-15).
You can select the target language, record some audio in English,
and then sit back and see how well the checkpoint can translate the input.
### Example
As this a standard sequence to sequence transformer model, you can use the `generate` method to generate the
transcripts by passing the speech features to the model.
You can use the model directly via the ASR pipeline. By default, the checkpoint will
translate spoken English to written German. To change the written target language,
you need to pass the correct `forced_bos_token_id` to `generate(...)` to condition
the decoder on the correct target language.
To select the correct `forced_bos_token_id` given your choosen language id, please make use
of the following mapping:
```python
MAPPING = {
"de": 250003,
"tr": 250023,
"fa": 250029,
"sv": 250042,
"mn": 250037,
"zh": 250025,
"cy": 250007,
"ca": 250005,
"sl": 250052,
"et": 250006,
"id": 250032,
"ar": 250001,
"ta": 250044,
"lv": 250017,
"ja": 250012,
}
```
As an example, if you would like to translate to Swedish, you can do the following:
```python
from datasets import load_dataset
from transformers import pipeline
# select correct `forced_bos_token_id`
forced_bos_token_id = MAPPING["sv"]
# replace following lines to load an audio file of your choice
librispeech_en = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation")
audio_file = librispeech_en[0]["file"]
asr = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-xls-r-1b-en-to-15", feature_extractor="facebook/wav2vec2-xls-r-1b-en-to-15")
translation = asr(audio_file, forced_bos_token_id=forced_bos_token_id)
```
or step-by-step as follows:
```python
import torch
from transformers import Speech2Text2Processor, SpeechEncoderDecoderModel
from datasets import load_dataset
model = SpeechEncoderDecoderModel.from_pretrained("facebook/wav2vec2-xls-r-1b-en-to-15")
processor = Speech2Text2Processor.from_pretrained("facebook/wav2vec2-xls-r-1b-en-to-15")
ds = load_dataset("patrickvonplaten/librispeech_asr_dummy", "clean", split="validation")
# select correct `forced_bos_token_id`
forced_bos_token_id = MAPPING["sv"]
inputs = processor(ds[0]["audio"]["array"], sampling_rate=ds[0]["audio"]["array"]["sampling_rate"], return_tensors="pt")
generated_ids = model.generate(input_ids=inputs["input_features"], attention_mask=inputs["attention_mask"], forced_bos_token_id=forced_bos_token)
transcription = processor.batch_decode(generated_ids)
```
## Results `en` -> `{lang}`
See the row of **XLS-R (1B)** for the performance on [Covost2](https://huggingface.co/datasets/covost2) for this model.
## More XLS-R models for `{lang}` -> `en` Speech Translation
- [Wav2Vec2-XLS-R-300M-EN-15](https://huggingface.co/facebook/wav2vec2-xls-r-300m-en-to-15)
- [Wav2Vec2-XLS-R-1B-EN-15](https://huggingface.co/facebook/wav2vec2-xls-r-1b-en-to-15)
- [Wav2Vec2-XLS-R-2B-EN-15](https://huggingface.co/facebook/wav2vec2-xls-r-2b-en-to-15)
- [Wav2Vec2-XLS-R-2B-22-16](https://huggingface.co/facebook/wav2vec2-xls-r-2b-22-to-16)
|