Merge pull request #182 from v3ucn/speed_change_sox_version

Speed change sox version

cosyvoice/utils/file_utils.py

@@ -39,3 +39,15 @@ def load_wav(wav, target_sr):
         assert sample_rate > target_sr, 'wav sample rate {} must be greater than {}'.format(sample_rate, target_sr)
         speech = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=target_sr)(speech)
     return speech
+
+def speed_change(waveform, sample_rate, speed_factor: str):
+    effects = [
+        ["tempo", speed_factor],  # speed_factor
+        ["rate", f"{sample_rate}"]
+    ]
+    augmented_waveform, new_sample_rate = torchaudio.sox_effects.apply_effects_tensor(
+        waveform,
+        sample_rate,
+        effects
+    )
+    return augmented_waveform, new_sample_rate

webui.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu)
+# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu, Liu Yue)
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -28,7 +28,7 @@ import logging
 logging.getLogger('matplotlib').setLevel(logging.WARNING)
 
 from cosyvoice.cli.cosyvoice import CosyVoice
-from cosyvoice.utils.file_utils import load_wav
+from cosyvoice.utils.file_utils import load_wav, speed_change
 
 logging.basicConfig(level=logging.DEBUG,
                     format='%(asctime)s %(levelname)s %(message)s')
@@ -66,7 +66,7 @@ instruct_dict = {'预训练音色': '1. 选择预训练音色\n2. 点击生成
 def change_instruction(mode_checkbox_group):
     return instruct_dict[mode_checkbox_group]
 
-def generate_audio(tts_text, mode_checkbox_group, sft_dropdown, prompt_text, prompt_wav_upload, prompt_wav_record, instruct_text, seed):
+def generate_audio(tts_text, mode_checkbox_group, sft_dropdown, prompt_text, prompt_wav_upload, prompt_wav_record, instruct_text, seed, speed_factor):
     if prompt_wav_upload is not None:
         prompt_wav = prompt_wav_upload
     elif prompt_wav_record is not None:
@@ -132,7 +132,16 @@ def generate_audio(tts_text, mode_checkbox_group, sft_dropdown, prompt_text, pro
         logging.info('get instruct inference request')
         set_all_random_seed(seed)
         output = cosyvoice.inference_instruct(tts_text, sft_dropdown, instruct_text)
-    audio_data = output['tts_speech'].numpy().flatten()
+    
+    if speed_factor != 1.0:
+        try:
+            audio_data, sample_rate = speed_change(output["tts_speech"], target_sr, str(speed_factor))
+            audio_data = audio_data.numpy().flatten()
+        except Exception as e:
+            print(f"Failed to change speed of audio: \n{e}")
+    else:
+        audio_data = output['tts_speech'].numpy().flatten()
+
     return (target_sr, audio_data)
 
 def main():
@@ -141,7 +150,7 @@ def main():
         gr.Markdown("#### 请输入需要合成的文本，选择推理模式，并按照提示步骤进行操作")
 
         tts_text = gr.Textbox(label="输入合成文本", lines=1, value="我是通义实验室语音团队全新推出的生成式语音大模型，提供舒适自然的语音合成能力。")
-
+        speed_factor = gr.Slider(minimum=0.25, maximum=4, step=0.05, label="语速调节", value=1.0, interactive=True)
         with gr.Row():
             mode_checkbox_group = gr.Radio(choices=inference_mode_list, label='选择推理模式', value=inference_mode_list[0])
             instruction_text = gr.Text(label="操作步骤", value=instruct_dict[inference_mode_list[0]], scale=0.5)
@@ -162,7 +171,7 @@ def main():
 
         seed_button.click(generate_seed, inputs=[], outputs=seed)
         generate_button.click(generate_audio,
-                              inputs=[tts_text, mode_checkbox_group, sft_dropdown, prompt_text, prompt_wav_upload, prompt_wav_record, instruct_text, seed],
+                              inputs=[tts_text, mode_checkbox_group, sft_dropdown, prompt_text, prompt_wav_upload, prompt_wav_record, instruct_text, seed, speed_factor],
                               outputs=[audio_output])
         mode_checkbox_group.change(fn=change_instruction, inputs=[mode_checkbox_group], outputs=[instruction_text])
     demo.queue(max_size=4, default_concurrency_limit=2)