def test_mfcc_extractor():
# get wave data
wavData = stream.read(wavPath).value
frames = stream.cut_frames(wavData)
# define an input pipe
pipe = base.PIPE()
for i in range(10):
pipe.put( base.Packet( {"rawWave":frames[i*50:(i+1)*50]}, cid=i, idmaker=0 ) )
pipe.stop()
print( pipe.size() )
# run a mfcc extractor
extractor = feature.MfccExtractor(oKey="mfcc")
extractor.start(inPIPE=pipe)
extractor.wait()
print( extractor.outPIPE.size() )
packet = extractor.outPIPE.get()
print( packet.keys() )
print( packet.mainKey )
print( packet["mfcc"].shape )
def test_processor_cmvn():
wavData = stream.read(wavPath).value
frames = stream.cut_frames(wavData)
pipe = base.PIPE()
for i in range(10):
pipe.put( base.Packet( {"rawWave":frames[i*50:(i+1)*50]}, cid=i, idmaker=0 ) )
pipe.stop()
extractor = feature.MfccExtractor(minParallelSize=100,oKey="mfcc")
processor = feature.MatrixFeatureProcessor(
spliceLeft=3,
spliceRight=3,
cmvNormalizer=feature.FrameSlideCMVNormalizer(),
oKey="mfcc",
)
extractor.start(inPIPE=pipe)
processor.start(inPIPE=extractor.outPIPE,iKey="mfcc")
processor.wait()
print( processor.outPIPE.size() )
packet = processor.outPIPE.get()
print( packet.keys() )
print( packet.mainKey )
print( packet["mfcc"].shape ) # 273 = 13 * 3 * 7
def test_mixture_feature_parallel():
wavData = stream.read(wavPath).value
frames = stream.cut_frames(wavData)
pipe = base.PIPE()
for i in range(10):
pipe.put(
base.Packet({"rawWave": frames[i * 50:(i + 1) * 50]},
cid=i,
idmaker=0))
pipe.stop()
extractor = feature.MixtureExtractor(mixType=["fbank", "mfcc"], )
# Split packets
def split_rule(items):
return {"fbank": items["fbank"]}, {"mfcc": items["mfcc"]}
spliter = joint.Spliter(split_rule, outNums=2)
# use a processor to transform fbank feature
processor1 = feature.MatrixFeatureProcessor(
spliceLeft=2,
spliceRight=2,
cmvNormalizer=feature.FrameSlideCMVNormalizer(),
oKey="fbank",
)
# use a processor to transform mfcc feature
processor2 = feature.MatrixFeatureProcessor(
spliceLeft=3,
spliceRight=3,
cmvNormalizer=feature.FrameSlideCMVNormalizer(),
oKey="mfcc",
)
# combine packets
def combine_rule(items):
return {
"feat": np.concatenate([items[0]["fbank"], items[1]["mfcc"]],
axis=1)
}
combiner = joint.Combiner(combine_rule)
extractor.start(inPIPE=pipe)
spliter.start(inPIPE=extractor.outPIPE)
processor1.start(
inPIPE=spliter.outPIPE[0]) # specify which key you want to process
processor2.start(
inPIPE=spliter.outPIPE[1]) # specify which key you want to process
combiner.start(inPIPE=[processor1.outPIPE, processor2.outPIPE])
combiner.wait()
print(combiner.outPIPE[0].size())
packet = combiner.outPIPE[0].get()
print(packet.keys())
print(packet["feat"].shape) # 211 = 120 + 91
def test_functions():
# Read wave info and data
wav = stream.read(wavPath)
print(wav.value)
# Cut the stream into N frames (discard the rest)
frames1 = stream.cut_frames(wav.value[:-10],
width=400,
shift=160,
snip=True)
print(frames1.shape)
# Cut the stream into N frames (retain the rest)
frames2 = stream.cut_frames(wav.value[:-10],
width=400,
shift=160,
snip=False)
print(frames2.shape)
def test_mixture_feature_series():
wavData = stream.read(wavPath).value
frames = stream.cut_frames(wavData)
pipe = base.PIPE()
for i in range(10):
pipe.put(
base.Packet({"rawWave": frames[i * 50:(i + 1) * 50]},
cid=i,
idmaker=0))
pipe.stop()
extractor = feature.MixtureExtractor(mixType=["fbank", "mfcc"], )
# use a processor to transform fbank feature
processor1 = feature.MatrixFeatureProcessor(
spliceLeft=2,
spliceRight=2,
cmvNormalizer=feature.FrameSlideCMVNormalizer(),
oKey="fbank",
)
# use a processor to transform mfcc feature
processor2 = feature.MatrixFeatureProcessor(
spliceLeft=3,
spliceRight=3,
cmvNormalizer=feature.FrameSlideCMVNormalizer(),
oKey="mfcc",
)
extractor.start(inPIPE=pipe)
processor1.start(inPIPE=extractor.outPIPE,
iKey="fbank") # specify which key you want to process
processor2.start(inPIPE=processor1.outPIPE,
iKey="mfcc") # specify which key you want to process
processor2.wait()
print(processor2.outPIPE.size())
packet = processor2.outPIPE.get()
print(packet.keys())
print(packet["fbank"].shape) # 120 = 24 * 5
print(packet["mfcc"].shape) # 91 = 13 * 7
