File size: 3,889 Bytes
3330d20
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
78c721b
 
 
 
6281348
78c721b
 
 
 
cce0d1e
78c721b
3330d20
 
 
8fef7e7
3330d20
 
 
 
8fef7e7
3330d20
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6281348
8fef7e7
3330d20
 
 
8fef7e7
78c721b
cce0d1e
3330d20
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
# Copyright      2022  Xiaomi Corp.        (authors: Fangjun Kuang)
#
# See LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from functools import lru_cache

from huggingface_hub import hf_hub_download


import sherpa_onnx
import numpy as np
from typing import Tuple
import wave

sample_rate = 16000


def read_wave(wave_filename: str) -> Tuple[np.ndarray, int]:
    """
    Args:
      wave_filename:
        Path to a wave file. It should be single channel and each sample should
        be 16-bit. Its sample rate does not need to be 16kHz.
    Returns:
      Return a tuple containing:
       - A 1-D array of dtype np.float32 containing the samples, which are
       normalized to the range [-1, 1].
       - sample rate of the wave file
    """

    with wave.open(wave_filename) as f:
        assert f.getnchannels() == 1, f.getnchannels()
        assert f.getsampwidth() == 2, f.getsampwidth()  # it is in bytes
        num_samples = f.getnframes()
        samples = f.readframes(num_samples)
        samples_int16 = np.frombuffer(samples, dtype=np.int16)
        samples_float32 = samples_int16.astype(np.float32)

        samples_float32 = samples_float32 / 32768
        return samples_float32, f.getframerate()


def decode(
    recognizer: sherpa_onnx.OfflineRecognizer,
    filename: str,
) -> str:
    s = recognizer.create_stream()
    samples, sample_rate = read_wave(filename)
    s.accept_waveform(sample_rate, samples)
    recognizer.decode_stream(s)

    return s.result.text.lower()


def _get_nn_model_filename(
    repo_id: str,
    filename: str,
    subfolder: str = ".",
) -> str:
    nn_model_filename = hf_hub_download(
        repo_id=repo_id,
        filename=filename,
        subfolder=subfolder,
    )
    return nn_model_filename


def _get_token_filename(
    repo_id: str,
    filename: str,
    subfolder: str = ".",
) -> str:
    token_filename = hf_hub_download(
        repo_id=repo_id,
        filename=filename,
        subfolder=subfolder,
    )
    return token_filename


@lru_cache(maxsize=8)
def get_pretrained_model(name: str) -> sherpa_onnx.OfflineRecognizer:
    assert name in (
        "tiny.en",
        "base.en",
        "small.en",
        "medium.en",
        "tiny",
        "base",
        "small",
        "medium",
        "medium-aishell",
    ), name
    full_repo_id = "csukuangfj/sherpa-onnx-whisper-" + name
    encoder = _get_nn_model_filename(
        repo_id=full_repo_id,
        filename=f"{name}-encoder.int8.onnx",
    )

    decoder = _get_nn_model_filename(
        repo_id=full_repo_id,
        filename=f"{name}-decoder.int8.onnx",
    )

    tokens = _get_token_filename(repo_id=full_repo_id, filename=f"{name}-tokens.txt")

    recognizer = sherpa_onnx.OfflineRecognizer.from_whisper(
        encoder=encoder,
        decoder=decoder,
        tokens=tokens,
        num_threads=2,
    )

    return recognizer


whisper_models = {
    "tiny.en": get_pretrained_model,
    "base.en": get_pretrained_model,
    "small.en": get_pretrained_model,
    "medium.en": get_pretrained_model,
    "distil-medium.en": get_pretrained_model,
    "tiny": get_pretrained_model,
    "base": get_pretrained_model,
    "small": get_pretrained_model,
    "distil-small.en": get_pretrained_model,
    "medium": get_pretrained_model,
    "medium-aishell": get_pretrained_model,
}