def decode_results(model, decoded_output, decoded_offsets):
results = {
"output": [],
"_meta": {
"acoustic_model": {
"name": os.path.basename(args.model_path)
},
"language_model": {
"name": os.path.basename(args.lm_path) if args.lm_path else None,
},
"decoder": {
"lm": args.lm_path is not None,
"alpha": args.alpha if args.lm_path is not None else None,
"beta": args.beta if args.lm_path is not None else None,
"type": args.decoder,
}
}
}
results['_meta']['acoustic_model'].update(DeepSpeech.get_meta(model))
for b in range(len(decoded_output)):
for pi in range(min(args.top_paths, len(decoded_output[b]))):
result = {'transcription': decoded_output[b][pi]}
if args.offsets:
result['offsets'] = decoded_offsets[b][pi].tolist()
results['output'].append(result)
return results
def decode_results(model, decoded_output, decoded_offsets):
results = {
"output": [],
"_meta": {
"acoustic_model": {
"name": os.path.basename(args.model_path)
},
"language_model": {
"name": os.path.basename(args.lm_path) if args.lm_path else None,
},
"decoder": {
"lm": args.lm_path is not None,
"alpha": args.alpha if args.lm_path is not None else None,
"beta": args.beta if args.lm_path is not None else None,
"type": args.decoder,
}
}
}
results['_meta']['acoustic_model'].update(DeepSpeech.get_meta(model))
for b in range(len(decoded_output)):
for pi in range(min(args.top_paths, len(decoded_output[b]))):
result = {'transcription': decoded_output[b][pi]}
if args.offsets:
result['offsets'] = decoded_offsets[b][pi]
results['output'].append(result)
return results
def decode_results(model, decoded_output, decoded_offsets):
results = {
"output": [],
"_meta": {
"acoustic_model": {
"name": os.path.basename(args.model_path)
},
"language_model": {
"name":
os.path.basename(args.lm_path) if args.lm_path else None,
},
"decoder": {
"lm": args.lm_path is not None,
"alpha": args.alpha if args.lm_path is not None else None,
"beta": args.beta if args.lm_path is not None else None,
"type": args.decoder,
}
}
}
results['_meta']['acoustic_model'].update(DeepSpeech.get_meta(model))
str = ''
print("len is : ", len(decoded_output))
for b in range(len(decoded_output)):
str2 = ''
for pi in range(min(args.top_paths, len(decoded_output[b]))):
result = {'transcription': decoded_output[b][pi]}
#if(decoded_output[b][pi]!=" "):
str2 += decoded_output[b][pi]
if args.offsets:
result['offsets'] = decoded_offsets[b][pi]
results['output'].append(result)
#str+=','
#str+=removerepeat(str2)
str += str2
str = removerepeat(str)
str = str.lower()
print(str)
#return results
return str
def word_decode(decoder, data, time_div=50, window=5, model=None):
strings, aligns, conf, char_probs = decoder.decode(data)
results = {
"one_best": "",
"num_paths": decoder._top_n,
"top_paths": [],
"_meta": {
"acoustic_model": {
"name": os.path.basename(args.model_path),
**DeepSpeech.get_meta(model)
},
"language_model": {
"name":
os.path.basename(args.lm_path) if args.lm_path else None,
"dict":
os.path.basename(args.dict_path) if args.dict_path else None
},
"decoder": {
"lm": args.lm_path is not None,
"dict": args.dict_path is not None,
"alpha": args.lm_alpha if args.lm_path is not None else None,
"beta": args.lm_beta if args.lm_path is not None else None,
"type": args.decoder,
"label_size": args.label_size,
"label_margin": args.label_margin
}
}
}
for pi in range(len(strings)):
for i in range(len(strings[pi])):
path = {"rank": pi, "conf": float(conf[pi][i]), "tokens": []}
word = ''
word_prob = 0.0
start_idx = -1
for idx, c in enumerate(strings[pi][i]):
if c == ' ' and word != '':
start_align = aligns[pi][i][start_idx]
end_align = aligns[pi][i][idx - 1] + window
path['tokens'].append({
"token": word,
"start": start_align / time_div,
"end": end_align / time_div,
"conf": word_prob
})
word = ''
start_idx = -1
else:
if start_idx == -1:
start_idx = idx
word += c
word_prob += char_probs[pi][i][idx]
if word != '':
path['tokens'].append({
"token":
word,
"start": (aligns[pi][i][start_idx]) / time_div,
"end": (aligns[pi][i][len(strings[pi][i]) - 1] + window) /
time_div,
"conf":
word_prob
})
results['top_paths'].append(path)
if len(results['top_paths']) > 0:
results['one_best'] = " ".join(
[x['token'] for x in results['top_paths'][0]['tokens']])
return results
