Function process_tts_multi

generate_infinitetalk.py:404–451

(text, save_dir, voice1, voice2)

   

def process_tts_multi(text, save_dir, voice1, voice2):
    pattern = r'\(s(\d+)\)\s*(.*?)(?=\s*\(s\d+\)|$)'
    matches = re.findall(pattern, text, re.DOTALL)
    
    s1_sentences = []
    s2_sentences = []

    pipeline = KPipeline(lang_code='a', repo_id='weights/Kokoro-82M')
    for idx, (speaker, content) in enumerate(matches):
        if speaker == '1':
            voice_tensor = torch.load(voice1, weights_only=True)
            generator = pipeline(
                content, voice=voice_tensor, # <= change voice here
                speed=1, split_pattern=r'\n+'
            )
            audios = []
            for i, (gs, ps, audio) in enumerate(generator):
                audios.append(audio)
            audios = torch.concat(audios, dim=0)
            s1_sentences.append(audios)
            s2_sentences.append(torch.zeros_like(audios))
        elif speaker == '2':
            voice_tensor = torch.load(voice2, weights_only=True)
            generator = pipeline(
                content, voice=voice_tensor, # <= change voice here
                speed=1, split_pattern=r'\n+'
            )
            audios = []
            for i, (gs, ps, audio) in enumerate(generator):
                audios.append(audio)
            audios = torch.concat(audios, dim=0)
            s2_sentences.append(audios)
            s1_sentences.append(torch.zeros_like(audios))
    
    s1_sentences = torch.concat(s1_sentences, dim=0)
    s2_sentences = torch.concat(s2_sentences, dim=0)
    sum_sentences = s1_sentences + s2_sentences
    save_path1 =f'{save_dir}/s1.wav'
    save_path2 =f'{save_dir}/s2.wav'
    save_path_sum = f'{save_dir}/sum.wav'
    sf.write(save_path1, s1_sentences, 24000) # save each audio file
    sf.write(save_path2, s2_sentences, 24000)
    sf.write(save_path_sum, sum_sentences, 24000)

    s1, _ = librosa.load(save_path1, sr=16000)
    s2, _ = librosa.load(save_path2, sr=16000)
    # sum, _ = librosa.load(save_path_sum, sr=16000)
    return s1, s2, save_path_sum

def generate(args):
    rank = int(os.getenv("RANK", 0))