Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from gtts import gTTS | |
| from pydub import AudioSegment | |
| from io import BytesIO | |
| import numpy as np | |
| import librosa | |
| def change_pitch_librosa(audio_segment, semitones): | |
| sr = audio_segment.frame_rate | |
| # 오디오 데이터를 numpy array로 변환 (float32) | |
| samples = np.array(audio_segment.get_array_of_samples()).astype(np.float32) | |
| samples /= 32768.0 # int16 범위를 [-1, 1]로 정규화 | |
| if audio_segment.channels > 1: | |
| # 다중 채널인 경우, (채널, 샘플수) 형태로 변환 | |
| samples = samples.reshape((-1, audio_segment.channels)).T | |
| else: | |
| samples = samples.flatten() | |
| # librosa의 pitch_shift 호출 시 keyword 인자로 전달 (sr와 n_steps) | |
| shifted = librosa.effects.pitch_shift(samples, sr=sr, n_steps=semitones) | |
| if audio_segment.channels > 1: | |
| shifted = shifted.T.flatten() | |
| else: | |
| shifted = shifted.flatten() | |
| shifted = np.clip(shifted, -1.0, 1.0) | |
| shifted_int16 = np.int16(shifted * 32767) | |
| new_audio = AudioSegment( | |
| shifted_int16.tobytes(), | |
| frame_rate=sr, | |
| sample_width=audio_segment.sample_width, | |
| channels=audio_segment.channels | |
| ) | |
| return new_audio | |
| def change_pitch(audio_segment, semitones): | |
| # 남성 목소리 (-10 semitones)에는 Librosa 방식을 사용하여 재생 속도를 유지합니다. | |
| if semitones == -10: | |
| return change_pitch_librosa(audio_segment, semitones) | |
| else: | |
| new_sample_rate = int(audio_segment.frame_rate * (2.0 ** (semitones / 12.0))) | |
| pitched_audio = audio_segment._spawn(audio_segment.raw_data, overrides={'frame_rate': new_sample_rate}) | |
| return pitched_audio.set_frame_rate(audio_segment.frame_rate) | |
| def alternating_tts(multiline_text): | |
| sentences = multiline_text.strip().splitlines() | |
| if len(sentences) > 10: | |
| sentences = sentences[:10] | |
| combined_audio = AudioSegment.silent(duration=0) | |
| for i, sentence in enumerate(sentences): | |
| sentence = sentence.strip() | |
| if not sentence: | |
| continue | |
| tts = gTTS(sentence, lang="bg") | |
| audio_file = BytesIO() | |
| tts.write_to_fp(audio_file) | |
| audio_file.seek(0) | |
| sentence_audio = AudioSegment.from_file(audio_file, format="mp3") | |
| if i % 2 == 0: | |
| sentence_audio = change_pitch(sentence_audio, -10) | |
| else: | |
| sentence_audio = change_pitch(sentence_audio, 3) | |
| combined_audio += sentence_audio + AudioSegment.silent(duration=500) | |
| output_file = "alternating_output.mp3" | |
| combined_audio.export(output_file, format="mp3") | |
| return output_file | |
| with gr.Blocks() as demo: | |
| gr.Markdown( | |
| "## Bulgarian TTS with Alternating Male and Female Voices\n" | |
| "최대 10문장을 입력하세요. 각 줄마다 하나의 문장을 입력하면, " | |
| "짝수번째 문장은 남성 목소리(피치 -10 반음, 정상 속도)로, " | |
| "홀수번째 문장은 여성 목소리(피치 +3 반음)로 생성됩니다." | |
| ) | |
| input_text = gr.Textbox( | |
| label="Enter sentences (max 10, one per line):", | |
| lines=10, | |
| placeholder="문장 1\n문장 2\n문장 3\n..." | |
| ) | |
| output_audio = gr.Audio(label="Generated Speech", type="filepath") | |
| generate_button = gr.Button("Generate Speech") | |
| generate_button.click(alternating_tts, inputs=input_text, outputs=output_audio) | |
| if __name__ == "__main__": | |
| demo.launch() | |