Spaces:
Runtime error
Runtime error
File size: 6,841 Bytes
80aa2c2 30e5a58 80aa2c2 30e5a58 e2fe5a6 30e5a58 cf75540 30e5a58 cf75540 a551fa6 cf75540 9c69c25 30e5a58 a551fa6 bf59269 28cac08 80aa2c2 3eaeb0a 339cfe2 82fbd24 339cfe2 82fbd24 e2fe5a6 30e5a58 cf75540 9c69c25 3654746 339cfe2 cf75540 339cfe2 e2fe5a6 339cfe2 cf75540 339cfe2 30e5a58 6cb3e95 3eaeb0a 784de0d 339cfe2 784de0d 339cfe2 30e5a58 82fbd24 5165246 bf59269 5165246 bf59269 82fbd24 30e5a58 82fbd24 30e5a58 6cb3e95 e2fe5a6 a551fa6 bf59269 61c73ae bf59269 30e5a58 88ec8bf 30e5a58 61c73ae bf59269 e031745 30e5a58 e2fe5a6 30e5a58 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 |
import logging
import sys
import gradio as gr
from transformers import pipeline, AutoModelForCTC, Wav2Vec2Processor, Wav2Vec2ProcessorWithLM
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
handlers=[logging.StreamHandler(sys.stdout)],
)
logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG)
LARGE_MODEL_BY_LANGUAGE = {
"Arabic": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-arabic", "has_lm": False},
"Chinese": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-chinese-zh-cn", "has_lm": False},
"Dutch": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-dutch", "has_lm": True},
"English": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-english", "has_lm": True},
"Finnish": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-finnish", "has_lm": False},
"French": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-french", "has_lm": True},
"German": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-german", "has_lm": True},
"Greek": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-greek", "has_lm": False},
"Hungarian": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-hungarian", "has_lm": False},
"Italian": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-italian", "has_lm": True},
"Japanese": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-japanese", "has_lm": False},
"Persian": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-persian", "has_lm": False},
"Polish": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-polish", "has_lm": True},
"Portuguese": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-portuguese", "has_lm": True},
"Russian": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-russian", "has_lm": True},
"Spanish": {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-spanish", "has_lm": True},
}
XLARGE_MODEL_BY_LANGUAGE = {
"Dutch": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-dutch", "has_lm": True},
"English": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-english", "has_lm": True},
"French": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-french", "has_lm": True},
"German": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-german", "has_lm": True},
"Italian": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-italian", "has_lm": True},
"Polish": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-polish", "has_lm": True},
"Portuguese": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-portuguese", "has_lm": True},
"Russian": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-russian", "has_lm": True},
"Spanish": {"model_id": "jonatasgrosman/wav2vec2-xls-r-1b-spanish", "has_lm": True},
}
# LANGUAGES = sorted(LARGE_MODEL_BY_LANGUAGE.keys())
# the container given by HF has 16GB of RAM, so we need to limit the number of models to load
LANGUAGES = sorted(XLARGE_MODEL_BY_LANGUAGE.keys())
CACHED_MODELS_BY_ID = {}
def run(input_file, language, decoding_type, history, model_size="300M"):
logger.info(f"Running ASR {language}-{model_size}-{decoding_type} for {input_file}")
# history = history or []
# the history seems to be not by session anymore, so I'll deactivate this for now
history = []
if model_size == "300M":
model = LARGE_MODEL_BY_LANGUAGE.get(language, None)
else:
model = XLARGE_MODEL_BY_LANGUAGE.get(language, None)
if model is None:
history.append({
"error_message": f"Model size {model_size} not found for {language} language :("
})
elif decoding_type == "LM" and not model["has_lm"]:
history.append({
"error_message": f"LM not available for {language} language :("
})
else:
# model_instance = AutoModelForCTC.from_pretrained(model["model_id"])
model_instance = CACHED_MODELS_BY_ID.get(model["model_id"], None)
if model_instance is None:
model_instance = AutoModelForCTC.from_pretrained(model["model_id"])
CACHED_MODELS_BY_ID[model["model_id"]] = model_instance
if decoding_type == "LM":
processor = Wav2Vec2ProcessorWithLM.from_pretrained(model["model_id"])
asr = pipeline("automatic-speech-recognition", model=model_instance, tokenizer=processor.tokenizer,
feature_extractor=processor.feature_extractor, decoder=processor.decoder)
else:
processor = Wav2Vec2Processor.from_pretrained(model["model_id"])
asr = pipeline("automatic-speech-recognition", model=model_instance, tokenizer=processor.tokenizer,
feature_extractor=processor.feature_extractor, decoder=None)
transcription = asr(input_file.name, chunk_length_s=5, stride_length_s=1)["text"]
logger.info(f"Transcription for {language}-{model_size}-{decoding_type} for {input_file}: {transcription}")
history.append({
"model_id": model["model_id"],
"language": language,
"model_size": model_size,
"decoding_type": decoding_type,
"transcription": transcription,
"error_message": None
})
html_output = "<div class='result'>"
for item in history:
if item["error_message"] is not None:
html_output += f"<div class='result_item result_item_error'>{item['error_message']}</div>"
else:
url_suffix = " + LM" if item["decoding_type"] == "LM" else ""
html_output += "<div class='result_item result_item_success'>"
html_output += f'<strong><a target="_blank" href="https://huggingface.co/{item["model_id"]}">{item["model_id"]}{url_suffix}</a></strong><br/><br/>'
html_output += f'{item["transcription"]}<br/>'
html_output += "</div>"
html_output += "</div>"
return html_output, history
gr.Interface(
run,
inputs=[
gr.inputs.Audio(source="microphone", type="file", label="Record something..."),
gr.inputs.Radio(label="Language", choices=LANGUAGES),
gr.inputs.Radio(label="Decoding type", choices=["greedy", "LM"]),
# gr.inputs.Radio(label="Model size", choices=["300M", "1B"]),
"state"
],
outputs=[
gr.outputs.HTML(label="Outputs"),
"state"
],
title="Automatic Speech Recognition",
description="",
css="""
.result {display:flex;flex-direction:column}
.result_item {padding:15px;margin-bottom:8px;border-radius:15px;width:100%}
.result_item_success {background-color:mediumaquamarine;color:white;align-self:start}
.result_item_error {background-color:#ff7070;color:white;align-self:start}
""",
allow_screenshot=False,
allow_flagging="never",
theme="grass"
).launch(enable_queue=True)
|