scripts/reaper2pass.py

import soundfile as sf
import numpy as np
from scipy import signal
from pydub import AudioSegment
import subprocess
import os

# 2 pass f0 estimation
# ref. Hirst The analysis by synthesis of speech melody: from data to models
# python wrap for gradio app 


# reaper requires wav file path input,
#  not audio data.
# 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}/files_tmp/{fname}tmp.wav'
    if not os.path.exists(f'{curdir}/files_tmp'):
        os.mkdir(f'{curdir}/files_tmp')
    aud_data.export(tmp_path, format="wav")
    wav_path = tmp_path
        
    return wav_path



# returns f0 data as list of Time, F0 if exists, voicing indicator
def get_reaper(wav_path, reaper_path, maxf0='700', minf0='50'):
    
    f0_data = subprocess.run([reaper_path, "-i", wav_path, '-f', '/dev/stdout', '-x', maxf0, '-m', minf0, '-a'],capture_output=True).stdout
    f0_data = f0_data.decode()
    f0_data = f0_data.split('EST_Header_End\n')[1].splitlines()
    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), float(v)] for t,v,f in f0_data]

    return f0_data
    
    

# save simplified data format from get_reaper
#  instead of reaper's original output
def save_pitch(f0_data, save_path,hed=False):
    with open(save_path,'w') as handle:
        if hed:
            handle.write('TIME\tF0\tVOICED\n')
        handle.write(''.join([f'{t}\t{f}\t{v}\n' for t,f,v in f0_data]))


# 2 pass pitch estimation
def estimate_pitch(sound_path,reaper_path = "REAPER/build/reaper"):

    orig_ftype = sound_path.split('.')[-1]
    if orig_ftype == 'wav':
        wav_path = sound_path
    else:
        tmp_path = reaper_soundfile(sound_path, orig_ftype)
        wav_path = tmp_path

    #print('REAPER FILE PATH:', wav_path)

    first_pass = get_reaper(wav_path,reaper_path)
    first_pass = [f for t,f,v in first_pass if float(v) ==1]
    
    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,reaper_path, maxf0 = str(round(pceil)), minf0 = str(round(pfloor)))


    #if orig_ftype != '.wav':
    #    subprocess.run(["rm", tmp_path])
    # don't remove it yet, need it for clustering too
    # therefore, actually change so reaper2pass is called from inside clusterprosody 
    # before it wants to read the f0 file.
    # TODO
        
    return second_pass