import click from loguru import logger import pydub from tqdm import tqdm logger.remove() logger.add(lambda msg: tqdm.write(msg, end=""), colorize=True) @click.group() def cli(): pass @cli.command() @click.option( "--backend", default="crispasr", type=click.Choice(["crispasr", "qtr"], case_sensitive=False), help="The backend to use for speech synthesis", ) @click.option( "--model", default="./qwen3tts", help="The model to use for speech recognition" ) @click.option( "--output", default="./output.wav", help="The output file for the generated speech" ) @click.option( "--output-dir", default="./outputs", help="The output directory for the generated speech", ) @click.option( "--gpu-backend", default="auto", help="The GPU backend to use for speech recognition", ) @click.option( "--instruct", default="用符合文本应读语气说", help="The instruction to use for speech recognition", ) @click.option( "executable", "--executable", default="./crispasr", help="The executable to use for speech recognition", ) @click.option( "--codec-model", help="The codec model to use for speech recognition", ) @click.option( "--ws-uri", default="ws://127.0.0.1:3001/api/tts/stream", help="The websocket uri used by qtr backend", ) @click.option( "--sample-rate", default=24000, type=int, help="Sample rate used by qtr backend when packaging wav", ) @click.option( "--strict", is_flag=True, help="Use strict parsing mode,which just keep Chinese, English and numbers", ) @click.option( "--split-length", default=450, type=int, help="The max length of each text chunk sent to TTS backend", ) @click.option( "--mode", default="mode", type=click.Choice(["stream", "http"], case_sensitive=False), help="The mode to use for speech synthesis", ) @click.option( "--form-chunk", help="Form chunk to say, 1 for first chunk, 2 for second chunk, etc.", ) @click.option( "--to-chunk", help="To chunk to say, 1 for first chunk, 2 for second chunk, etc.", ) @click.argument("input_file") def say( backend: str, model: str, output: str, input_file: str, gpu_backend: str, instruct: str, executable: str, output_dir: str, codec_model: str, ws_uri: str, sample_rate: int, strict: bool, mode: str, split_length: int, form_chunk: int, to_chunk: int, ): with open(input_file, "r", encoding="utf-8") as f: html = f.read() from parser import TTSHTMLParser logger.info("解析中...") parser = TTSHTMLParser(html) result = parser.parse(strict=strict) logger.debug(f"解析结果: {result}") match backend: case "crispasr": from backend import CrispASRCLIBackend tts_backend = CrispASRCLIBackend( model=model, executable=executable, gpu_backend=gpu_backend, codec_model=codec_model, ) case "qtr": from backend import QTRTTSBackend tts_backend = QTRTTSBackend(ws_uri=ws_uri, sample_rate=sample_rate) from tqdm import tqdm case _: raise ValueError(f"Unsupported backend: {backend}") for voice, text in tqdm( result, desc="合成语音", unit="段", total=len(result), colour="green", ): logger.info(f"正在合成: voice={voice}, text={text}", mode=mode) u tts_backend.smart_say( from_chunk=form_chunk, to_chunk=to_chunk, voice=voice, text=text, instruct=instruct, split_length=split_length, output_file_prefix=f"{output_dir}/{voice}_{text[0:9]}", ) @cli.command() @click.argument("input_dir") @click.argument("output_file") @click.option( "--add-silence", default=0.5, type=float, help="Add silence between combined wav files in seconds", ) @click.option("--format", default="wav", help="The output format, either wav or mp3") def combine(input_dir: str, output_file: str, add_silence: float, format: str): from files import FileHandler handler = FileHandler(input_dir) if format == "wav": handler.combine_waves(output_file, add_silence=add_silence) elif format == "mp3": handler.combine_waves(output_file) handler.wavs_to_mp3(output_file, output_file.replace(".wav", ".mp3")) @cli.command() @click.argument("input_file") @click.argument("output_file") def transform(input_file: str, output_file: str): from files import FileHandler handler = FileHandler(input_file) handler.wavs_to_mp3(input_file, output_file) @cli.command() @click.argument("input_file") @click.option( "--bgm-dir", default="./bgms", help="The directory containing background music files", ) @click.option( "--bgm-volume", default=-5, type=int, help="The volume of the background music in dB", ) @click.option( "--audio-refer", default="input_file", help="The audio file to use as reference for the final output", ) @click.option( "--mp3-bitrate", default="64k", help="Bitrate used when output file is mp3, e.g. 64k/96k/128k", ) @click.option( "--output-channels", default="keep", type=click.Choice(["keep", "mono", "stereo"], case_sensitive=False), help="Output channels: keep original, force mono, or force stereo", ) @click.option( "--output-sample-rate", default=0, type=int, help="Output sample rate in Hz, use 0 to keep original", ) @click.argument("output_file") def add_bgms( input_file: str, bgm_dir: str, bgm_volume: int, audio_refer: str, mp3_bitrate: str, output_channels: str, output_sample_rate: int, output_file: str, ): from pathlib import Path suffix = output_file.split(".")[-1] if suffix not in ["wav", "mp3"]: raise ValueError("Output file must have .wav or .mp3 suffix") match suffix: case "wav": main_audio = pydub.AudioSegment.from_wav(input_file) case "mp3": main_audio = pydub.AudioSegment.from_mp3(input_file) bgm_files = sorted( [ f for f in Path(bgm_dir).glob("*") if f.suffix.lower() in [".wav", ".mp3", ".ogg"] ], key=lambda f: f.stat().st_mtime, ) if not bgm_files: raise ValueError("No supported bgm files found in bgm directory") for bgm_file in bgm_files: bgm_suffix = bgm_file.suffix.lower() match bgm_suffix: case ".wav": bgm_audio = pydub.AudioSegment.from_wav(bgm_file) case ".mp3": bgm_audio = pydub.AudioSegment.from_mp3(bgm_file) case ".ogg": bgm_audio = pydub.AudioSegment.from_ogg(bgm_file) case _: logger.warning( f"Unsupported bgm format: {bgm_suffix}, skipping {bgm_file}" ) continue bgm_audio = bgm_audio - (20 - bgm_volume) # Adjust volume if len(bgm_audio) < len(main_audio): bgm_audio = bgm_audio * (len(main_audio) // len(bgm_audio) + 1) bgm_audio = bgm_audio[: len(main_audio)] match audio_refer: case "input_file": main_audio = main_audio.overlay(bgm_audio) case "bgm": main_audio = bgm_audio.overlay(main_audio) case _: raise ValueError(f"Unsupported audio refer: {audio_refer}") match output_channels.lower(): case "mono": main_audio = main_audio.set_channels(1) case "stereo": main_audio = main_audio.set_channels(2) case "keep": pass case _: raise ValueError(f"Unsupported output channels option: {output_channels}") if output_sample_rate > 0: main_audio = main_audio.set_frame_rate(output_sample_rate) if suffix == "mp3": main_audio.export(output_file, format="mp3", bitrate=mp3_bitrate) else: main_audio.export(output_file, format="wav") @cli.command() @click.option( "--remove-silence", default=0.3, type=float, help="Remove silence from the beginning and end of the audio", ) @click.argument("input_file") @click.argument("output_file") def pretty(remove_silence: float, input_file: str, output_file: str): if remove_silence > 0: from pretty import remove_silence as remove_silence_func remove_silence_func( pydub.AudioSegment.from_wav(input_file), remove_silence ).export(output_file, format="wav") @cli.command() @click.argument("input_txt") @click.argument("output_xml") @click.option( "--model", default="qwen3:latest", help="The model to use for text transformation, e.g. qwen3:4b or qwen3:latest", ) @click.option( "--context-window", default=2048, type=int, help="The context window size in characters for the text transformation model", ) @click.option( "--prompt", default=None, help="The prompt to use for text transformation, if not provided, a default prompt will be used", ) @click.option( "--speakers", default=None, help="Comma-separated list of speakers to use for text transformation, if not provided, a default list will be used", type=list, ) @click.option( "--transformer", default="TXTTransformer", type=click.Choice(["TXTTransformer", "OllamaTransformer"]), ) def txt_to_xml( input_txt: str, output_xml: str, model: str, context_window: int, prompt: str, speakers: list[str], transformer: str, ): from transform import TXTTransformer, OllamaTransformer, AutoTransformer with open(input_txt, "r", encoding="utf-8") as f: text = f.read() if transformer == "OllamaTransformer": transformer = OllamaTransformer( model=model, context_window=context_window, prompt=prompt, speakers=speakers ) elif transformer == "AutoTransformer": transformer = AutoTransformer(speakers=speakers) else: transformer = TXTTransformer() xml = transformer.transform(text) with open(output_xml, "w", encoding="utf-8") as f: f.write(xml)