准确的音素识别不容易存档,因为音素本身的定义相当松散。即使在好的音频中,现在最好的系统也有18%的音素错误率(你可以在Alex Graves发布的TIMIT上查看LSTM-RNN结果)。
在cmusphenx中,Python中的音素识别是这样完成的:from os import environ, path
from pocketsphinx.pocketsphinx import *
from sphinxbase.sphinxbase import *
MODELDIR = "../../../model"
DATADIR = "../../../test/data"
# Create a decoder with certain model
config = Decoder.default_config()
config.set_string('-hmm', path.join(MODELDIR, 'en-us/en-us'))
config.set_string('-allphone', path.join(MODELDIR, 'en-us/en-us-phone.lm.dmp'))
config.set_float('-lw', 2.0)
config.set_float('-beam', 1e-10)
config.set_float('-pbeam', 1e-10)
# Decode streaming data.
decoder = Decoder(config)
decoder.start_utt()
stream = open(path.join(DATADIR, 'goforward.raw'), 'rb')
while True:
buf = stream.read(1024)
if buf:
decoder.process_raw(buf, False, False)
else:
break
decoder.end_utt()
hypothesis = decoder.hyp()
print ('Phonemes: ', [seg.word for seg in decoder.seg()])
为了运行这个示例,您需要从github签出最新的pocketsphinx。结果应该是这样的:('Best phonemes: ', ['SIL', 'G', 'OW', 'F', 'AO', 'R', 'W', 'ER', 'D', 'T', 'AE', 'N', 'NG', 'IY', 'IH', 'ZH', 'ER', 'Z', 'S', 'V', 'SIL'])