app

scripts/reaper2pass.py

@@ -0,0 +1,73 @@
+import soundfile as sf
+import numpy as np
+from scipy import signal
+from pydub import AudioSegment
+import subprocess
+import os
+
+# ref. Hirst The analysis by synthesis of speech melody: from data to models
+
+
+# reaper requires wav file path input,
+#  not audio data itself.
+# reaper does NOT require 16khz mono audio.
+def reaper_soundfile(sound_path, orig_filetype):
+    
+    aud_data = AudioSegment.from_file(sound_path, orig_filetype)
+    curdir = subprocess.run(["pwd"], capture_output=True, text=True)
+    curdir = curdir.stdout.splitlines()[0]
+    fname = sound_path.split('/')[-1].replace(orig_filetype,'')
+    tmp_path = f'{curdir}/REAPER_TMP/{fname}_tmp.wav'
+    if not os.path.exists(f'{curdir}/REAPER_TMP'):
+        os.mkdir(f'{curdir}/REAPER_TMP')
+    aud_data.export(tmp_path, format="wav")
+    wav_path = tmp_path
+        
+    return wav_path
+
+
+
+def get_reaper(wav_path, maxf0='700', minf0='50', reaper_path = "REAPER/build/reaper"):
+    
+    f0_data = subprocess.run([reaper_path, "-i", wav_path, '-f', '/dev/stdout', '-x', maxf0, '-m', minf0, '-a'],capture_output=True).stdout
+    #print('PLAIN:',f0_data)
+    f0_data = f0_data.decode()
+    #print('DECODE-PITCH:',f0_data)
+    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 = [l for l in f0_data if len(l) == 3] # the last line or 2 lines are other info, different format
+    f0_data = [ [float(t), float(f)] for t,v,f in f0_data if v=='1']
+
+    return f0_data
+    
+
+
+# 2 pass pitch estimation
+def estimate_pitch(sound_path):
+
+    orig_ftype = sound_path.split('.')[-1]
+    if orig_ftype == '.wav':
+        wav_path = sound_path
+    else:
+        tmp_path = reaper_soundfile(sound_path)
+        wav_path = tmp_path
+
+    print('REAPER FILE PATH:', wav_path)
+
+    first_pass = get_reaper(wav_path)
+    first_pass = [f for t,f in first_pass]
+    
+    q1 = np.quantile(first_pass,0.25)
+    q3 = np.quantile(first_pass,0.75)
+    
+    pfloor = 0.75 * q1
+    pceil = 1.5 * q3
+
+    second_pass = get_reaper(wav_path,maxf0 = str(round(pceil)), minf0 = str(round(pfloor)))
+
+
+    if orig_ftype != '.wav':
+        subprocess.run(["rm", tmp_path])
+        
+    return second_pass

scripts/runSQ.py

@@ -1,6 +1,7 @@
 import os, unicodedata
 from scripts.ctcalign import aligner, wav16m
 from scripts.tapi import tiro
+from scripts.reaper2pass import estimate_pitch
 
 # given a Sentence string,
 # using a metadata file of SQ, // SQL1adult_metadata.tsv
@@ -34,6 +35,14 @@ def run(sentence, voices):
         f0_tts(sentence, voices, tts_dir, 'TODO path to reaper')
 
     # by now, all the data to cluster and eval exists in the right place.
+    # (after the last todo of saving pitch to disk instead of only list)
+
+    # next, make a thing that does clustering.
+    # its input is Meta + the paths to find wav, aln, f0 datas.
+
+    # its output may as well actually be graphs lol
+
+    # also stop forgetting duration.
         
     return temp_a_sample
 
@@ -115,7 +124,17 @@ def f0_human(meta, f0_dir, speech_dir, reaper_path):
         print(f'Need to estimate pitch for {len(no_f0)} recordings')
         if not os.path.exists(f0_dir):
             os.makedirs(f0_dir)
+        for rec in no_f0:
+            wav_path = f'{speech_dir}{rec[2]}'
+            temp_data_f0 = estimate_pitch(wav_path)
+            print('2ND PASS PITCHES FOR', wav_path)
+            print(temp_data_f0)
+
         #TODO
+        # Current Todo:
+        # have pitch saved to file instead of returned to data
+        # tbqh can write my own simplified files instead of make reaper write
+        #whatever.
         
     else:
         print('All speech pitch trackings existed')