import os import gradio as gr import requests import langid import base64 import json import time import re import hashlib import hash_code_for_cached_output API_URL = os.environ.get("API_URL") TOKEN = os.environ.get("TOKEN") RESULT_URL = os.environ.get("RESULT_URL") supported_languages = ['zh', 'en', 'ja', 'ko', 'es', 'fr'] supported_styles = { 'zh': "zh_default", 'en': [ "en_default", "en_us", "en_br", "en_au", "en_in" ], "es": "es_default", "fr": "fr_default", "ja": "jp_default", "ko": "kr_default" } output_dir = 'outputs' os.makedirs(output_dir, exist_ok=True) def audio_to_base64(audio_file): with open(audio_file, "rb") as audio_file: audio_data = audio_file.read() base64_data = base64.b64encode(audio_data).decode("utf-8") return base64_data def count_chars_words(sentence): segments = re.findall(r'[\u4e00-\u9fa5]+|\w+', sentence) char_count = 0 word_count = 0 for segment in segments: if re.match(r'[\u4e00-\u9fa5]+', segment): char_count += len(segment) else: word_count += len(segment.split()) return char_count + word_count def predict(prompt, style, audio_file_pth, speed, agree): # initialize a empty info text_hint = '' # agree with the terms if agree == False: text_hint += '[ERROR] Please accept the Terms & Condition!\n' gr.Warning("Please accept the Terms & Condition!") return ( text_hint, None, None, ) # Before we get into inference, we will detect if it is from example table or default value # If so, we use a cached Audio. Noted that, it is just for demo efficiency. # hash code were generated by `hash_code_for_cached_output.py` # this hash get from gradio console cached_outputs = { "af39e1f1ff_60565a5c20_en_us" : "cached_outputs/0.wav", "af39e1f1ff_420ab8211d_en_us" : "cached_outputs/1.wav", "ced034cc22_0f96bf44f5_es_default" : "cached_outputs/2.wav", "d3172b178d_3fef5adc6f_zh_default" : "cached_outputs/3.wav", "cda6998e1a_9897b60a4e_jp_default" : "cached_outputs/4.wav" } unique_code = hash_code_for_cached_output.get_unique_code(audio_file_pth, prompt, style) print("audio_file_pth is", audio_file_pth) print("unique_code is", unique_code) if unique_code in list(cached_outputs.keys()): return ( 'We get the cached output for you, since you are trying to generate an example cloning.', cached_outputs[unique_code], audio_file_pth, ) # first detect the input language language_predicted = langid.classify(prompt)[0].strip() print(f"Detected language:{language_predicted}") if language_predicted not in supported_languages: text_hint += f"[ERROR] The detected language {language_predicted} for your input text is not in our Supported Languages: {supported_languages}\n" gr.Warning( f"The detected language {language_predicted} for your input text is not in our Supported Languages: {supported_languages}" ) return ( text_hint, None, None, ) # check the style if style not in supported_styles[language_predicted]: text_hint += f"[Warming] The style {style} is not supported for detected language {language_predicted}. For language {language_predicted}, we support styles: {supported_styles[language_predicted]}. Using the wrong style may result in unexpected behavior.\n" gr.Warning(f"[Warming] The style {style} is not supported for detected language {language_predicted}. For language {language_predicted}, we support styles: {supported_styles[language_predicted]}. Using the wrong style may result in unexpected behavior.") prompt_length = count_chars_words(prompt) speaker_wav = audio_file_pth if prompt_length < 2: text_hint += f"[ERROR] Please give a longer prompt text \n" gr.Warning("Please give a longer prompt text") return ( text_hint, None, None, ) if prompt_length > 50: text_hint += f"[ERROR] Text length limited to 50 words for this demo, please try shorter text. You can clone our open-source repo or try it on our website https://app.myshell.ai/robot-workshop/widget/174760057433406749 \n" gr.Warning( "Text length limited to 50 words for this demo, please try shorter text. You can clone our open-source repo or try it on our website https://app.myshell.ai/robot-workshop/widget/174760057433406749" ) return ( text_hint, None, None, ) save_path = f'{output_dir}/output.wav' speaker_audio_base64 = audio_to_base64(speaker_wav) if style == 'en_us': # we update us accent style = 'en_newest' data = { "text": prompt, "reference_speaker": speaker_audio_base64, "language": style, "speed": speed } start = time.time() headers = { "Authorization": f"Bearer {TOKEN}" } response = requests.post(API_URL, json=data, headers=headers, timeout=60) print(f'Get response successfully within {time.time() - start}') task_id = response.json()['task_id'] while True: response = requests.post(RESULT_URL, json={'task_id': task_id}, headers=headers) json_data = response.json() status = json_data['status'] if status in ["CREATED", "RUNNING"]: time.sleep(1) continue if status == 'FAILED': text_hint += f"[HTTP ERROR] {json_data['error']} \n" gr.Warning( f"[HTTP ERROR] {json_data['error']} \n" ) return ( text_hint, None, None, ) else: decoded_bytes = base64.b64decode(json_data['result']['base64'].encode('utf-8')) with open(save_path, 'wb') as f: f.write(decoded_bytes) text_hint += f'''Get response successfully \n''' return ( text_hint, save_path, speaker_wav, ) title = "MyShell OpenVoice V2" description = """ In December 2023, we released [OpenVoice V1](https://huggingface.co/spaces/myshell-ai/OpenVoice), an instant voice cloning approach that replicates a speaker's voice and generates speech in multiple languages using only a short audio clip. OpenVoice V1 enables granular control over voice styles, replicates the tone color of the reference speaker and achieves zero-shot cross-lingual voice cloning. """ description_v2 = """ In April 2024, we released **OpenVoice V2**, which includes all features in V1 and has: - **Better Audio Quality**. OpenVoice V2 adopts a different training strategy that delivers better audio quality. - **Native Multi-lingual Support**. English, Spanish, French, Chinese, Japanese and Korean are natively supported in OpenVoice V2. - **Free Commercial Use**. Starting from April 2024, both V2 and V1 are released under MIT License. Free for commercial use. """ markdown_table = """
| | | | | :-----------: | :-----------: | :-----------: | | **OpenSource Repo** | **Project Page** | **Join the Community** | |
| [OpenVoice](https://research.myshell.ai/open-voice) | [![Discord](https://img.shields.io/discord/1122227993805336617?color=%239B59B6&label=%20Discord%20)](https://discord.gg/myshell) |
""" markdown_table_v2 = """
| | | | | | :-----------: | :-----------: | :-----------: | :-----------: | | **Github Repo** |
| **Project Page** | [OpenVoice](https://research.myshell.ai/open-voice) | | | | | :-----------: | :-----------: | **Join the Community** | [![Discord](https://img.shields.io/discord/1122227993805336617?color=%239B59B6&label=%20Discord%20)](https://discord.gg/myshell) |
""" content = """
If the generated voice does not sound like the reference voice, please refer to this QnA. If you want to deploy the model by yourself and perform inference, please refer to this jupyter notebook.
""" wrapped_markdown_content = f"
{content}
" examples = [ [ "Did you ever hear a folk tale about a giant turtle?", 'en_us', "examples/speaker0.mp3", True, ],[ "El resplandor del sol acaricia las olas, pintando el cielo con una paleta deslumbrante.", 'es_default', "examples/speaker1.mp3", True, ],[ "我最近在学习machine learning，希望能够在未来的artificial intelligence领域有所建树。", 'zh_default', "examples/speaker2.mp3", True, ],[ "彼は毎朝ジョギングをして体を健康に保っています。", 'jp_default', "examples/speaker3.mp3", True, ], ] with gr.Blocks(analytics_enabled=False) as demo: with gr.Row(): with gr.Column(): with gr.Row(): gr.Markdown( """ ## """ ) with gr.Row(): gr.Markdown(markdown_table_v2) with gr.Row(): gr.Markdown(description) with gr.Column(): gr.Video('./openvoicev2.mp4', autoplay=True) with gr.Row(): gr.Markdown(description_v2) with gr.Row(): gr.HTML(wrapped_markdown_content) with gr.Row(): with gr.Column(): input_text_gr = gr.Textbox( label="Text Prompt", info="One or two sentences at a time is better. Up to 200 text characters.", value="The bustling city square bustled with street performers, tourists, and local vendors.", ) style_gr = gr.Dropdown( label="Style", info="Select a style of output audio for the synthesised speech. (Chinese only support 'default' now)", choices=["en_default", "en_us", "en_br", "en_au", "en_in", "es_default", "fr_default", "jp_default", "zh_default", "kr_default",], max_choices=1, value="en_us", ) ref_gr = gr.Audio( label="Reference Audio", info="Click on the ✎ button to upload your own target speaker audio", type="filepath", value="examples/speaker0.mp3", ) tos_gr = gr.Checkbox( label="Agree", value=False, info="I agree to the terms of the MIT license-: https://github.com/myshell-ai/OpenVoice/blob/main/LICENSE", ) tts_button = gr.Button("Send", elem_id="send-btn", visible=True) with gr.Column(): out_text_gr = gr.Text(label="Info") audio_gr = gr.Audio(label="Synthesised Audio", autoplay=True) ref_audio_gr = gr.Audio(label="Reference Audio Used") gr.Examples(examples, label="Examples", inputs=[input_text_gr, style_gr, ref_gr, tos_gr], outputs=[out_text_gr, audio_gr, ref_audio_gr], fn=predict, cache_examples=False,) tts_button.click(predict, [input_text_gr, style_gr, ref_gr, tos_gr], outputs=[out_text_gr, audio_gr, ref_audio_gr]) demo.queue(concurrency_count=6) demo.launch(debug=True, show_api=True)