File size: 3,090 Bytes
459923a 424bfb6 459923a 424bfb6 459923a |
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 |
import numpy as np
import soundfile as sf
from scipy import signal
import librosa
import subprocess
import matplotlib.pyplot as plt
def readwav(wav_path):
wav, sr = sf.read(wav_path, dtype=np.float32)
if len(wav.shape) == 2:
wav = wav.mean(1)
if sr != 16000:
wlen = int(wav.shape[0] / sr * 16000)
wav = signal.resample(wav, wlen)
return wav
def normalise_transcript(xcp):
xcp = xcp.lower()
while ' ' in xcp:
xcp = xcp.replace(' ', ' ')
return xcp
def get_pitch_tracks(wav_path):
f0_data = subprocess.run(["REAPER/build/reaper", "-i", wav_path, '-a']).stdout
#with open('tmp.f0','r') as handle:
f0_data = f0_data.split('EST_Header_End\n')[1].splitlines()
print(f0_data) #!!!!!!!!!!!!!!!!!!!!!
f0_data = [l.split(' ') for l in f0_data]
f0_data = [ [float(t), float(f)] for t,v,f in f0_data if v=='1']
f0_data = [[t,f0] for t,prob,f0 in f0_data if prob==1.0]
return f0_data
# transcript could be from a corpus with the wav file,
# input by the user,
# or from a previous speech recognition process
def align_and_graph(wav_path, transcript, aligner_function):
# fetch data
speech = readwav(wav_path)
w_align, seg_align = aligner_function(speech,normalise_transcript(transcript))
# set up the graph shape
rec_start = w_align[0][1]
rec_end = w_align[-1][2]
f0_data = get_pitch_tracks(wav_path)
if f0_data:
f_max = max([f0 for t,f0 in f0_data]) + 50
else:
f_max = 400
fig, axes1 = plt.subplots(figsize=(15,5))
plt.xlim([rec_start, rec_end])
axes1.set_ylim([0.0, f_max])
axes1.get_xaxis().set_visible(False)
# draw word boundaries
for w,s,e in w_align:
plt.vlines(s,0,f_max,linewidth=0.5,color='black')
plt.vlines(e,0,f_max,linewidth=0.5,color='black')
plt.text( (s+e)/2 - (len(w)*.01), f_max+15, w, fontsize=15)
# draw phone/char boundaries
for p,s,e in seg_align:
plt.vlines(s,0,f_max,linewidth=0.3,color='cadetblue',linestyle=(0,(10,4)))
plt.vlines(e,0,f_max,linewidth=0.3,color='cadetblue',linestyle=(0,(10,4)))
plt.text( (s+e)/2 - (len(p)*.01), -30, p, fontsize=15, color='teal')
f0c = "blue"
axes1.scatter([t for t,f0 in f0_data], [f0 for t,f0 in f0_data], color=f0c)
w, sr = librosa.load(wav_path)
fr_l = 2048 # librosa default
h_l = 512 # default
rmse = librosa.feature.rms(y=w, frame_length = fr_l, hop_length = h_l)
rmse = rmse[0]
# show rms energy
axes2 = axes1.twinx()
axes2.set_ylim([0.0, 0.5])
rms_xval = [(h_l*i)/sr for i in range(len(rmse))]
axes2.plot(rms_xval,rmse,color='peachpuff',linewidth=3.5)
# label the graph
axes1.set_ylabel("Pitch (F0, Hz)", fontsize=14, color="blue")
axes2.set_ylabel("RMS energy", fontsize=14,color="coral")
#plt.title(f'Recording {file_id} (L1 {language_dict[file_id]})', fontsize=15)
#plt.show()
return fig
#plt.close('all')
# uppboðssøla bussleiðini viðmerkingar upprunaligur
|