From 8182908f7d147383705f7c2199ac126d9e24b0ba Mon Sep 17 00:00:00 2001 From: Downupanddownup Date: Mon, 29 Apr 2024 14:14:16 +0800 Subject: [PATCH] =?UTF-8?q?=E6=B7=BB=E5=8A=A0=E8=AF=B4=E8=AF=9D=E4=BA=BA?= =?UTF-8?q?=E7=A1=AE=E8=AE=A4=E6=A8=A1=E5=9E=8B=E5=88=87=E6=8D=A2?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- Ref_Audio_Selector/common/model_manager.py | 12 +++++++ .../ref_audio_selector_webui.py | 36 +++++++++++++------ .../speaker_verification/voice_similarity.py | 23 ++++++++---- 3 files changed, 55 insertions(+), 16 deletions(-) diff --git a/Ref_Audio_Selector/common/model_manager.py b/Ref_Audio_Selector/common/model_manager.py index ac2e7fb..b0f43fb 100644 --- a/Ref_Audio_Selector/common/model_manager.py +++ b/Ref_Audio_Selector/common/model_manager.py @@ -8,6 +8,18 @@ GPT_weight_root = "GPT_weights" os.makedirs(SoVITS_weight_root, exist_ok=True) os.makedirs(GPT_weight_root, exist_ok=True) +speaker_verification_models = { + 'speech_campplus_sv_zh-cn_16k-common': { + 'task': 'speaker-verification', + 'model': 'Ref_Audio_Selector/tool/speaker_verification/models/speech_campplus_sv_zh-cn_16k-common', + 'model_revision': 'v1.0.0' + }, + 'speech_eres2net_sv_zh-cn_16k-common': { + 'task': 'speaker-verification', + 'model': 'Ref_Audio_Selector/tool/speaker_verification/models/speech_eres2net_sv_zh-cn_16k-common', + 'model_revision': 'v1.0.5' + } +} def custom_sort_key(s): # 使用正则表达式提取字符串中的数字部分和非数字部分 diff --git a/Ref_Audio_Selector/ref_audio_selector_webui.py b/Ref_Audio_Selector/ref_audio_selector_webui.py index 0e22ace..73579b6 100644 --- a/Ref_Audio_Selector/ref_audio_selector_webui.py +++ b/Ref_Audio_Selector/ref_audio_selector_webui.py @@ -71,7 +71,7 @@ def convert_from_list(text_work_space_dir, text_role, text_list_input): return i18n(text_convert_from_list_info), text_sample_dir -def start_similarity_analysis(work_space_dir, sample_dir, base_voice_path, need_similarity_output): +def start_similarity_analysis(work_space_dir, sample_dir, speaker_verification, base_voice_path, need_similarity_output): similarity_list = None similarity_file_dir = None @@ -87,6 +87,7 @@ def start_similarity_analysis(work_space_dir, sample_dir, base_voice_path, need_ cmd += f' -r "{base_voice_path}"' cmd += f' -c "{sample_dir}"' cmd += f' -o {similarity_file}' + cmd += f' -m {speaker_verification}' logger.info(cmd) p_similarity = Popen(cmd, shell=True) @@ -105,7 +106,7 @@ def start_similarity_analysis(work_space_dir, sample_dir, base_voice_path, need_ # 基于一个基准音频，从参考音频目录中进行分段抽样 -def sample(text_work_space_dir, text_role, text_sample_dir, text_base_voice_path, +def sample(text_work_space_dir, text_role, text_sample_dir, dropdown_speaker_verification, text_base_voice_path, slider_subsection_num, slider_sample_num, checkbox_similarity_output): text_work_space_dir, text_sample_dir, text_base_voice_path \ = common.batch_clean_paths([text_work_space_dir, text_sample_dir, text_base_voice_path]) @@ -122,11 +123,13 @@ def sample(text_work_space_dir, text_role, text_sample_dir, text_base_voice_path raise Exception("分段数不能为空") if slider_sample_num is None or slider_sample_num == '': raise Exception("每段随机抽样个数不能为空") + if dropdown_speaker_verification is None or dropdown_speaker_verification == '': + raise Exception("说话人确认算法不能为空") ref_audio_dir = os.path.join(base_role_dir, params.reference_audio_dir) time_consuming, (similarity_list, _, _) \ - = time_util.time_monitor(start_similarity_analysis)(base_role_dir, text_sample_dir, text_base_voice_path, + = time_util.time_monitor(start_similarity_analysis)(base_role_dir, text_sample_dir, dropdown_speaker_verification, text_base_voice_path, checkbox_similarity_output) text_sample_info = f"耗时：{time_consuming:0.1f}秒；抽样成功：生成目录{ref_audio_dir}" @@ -315,7 +318,7 @@ hide_voice_similarity_dir = '' # 根据一个参考音频，对指定目录下的音频进行相似度分析，并输出到另一个目录 def similarity_audio_output(text_work_space_dir, text_role, text_base_audio_path, - text_compare_audio_dir): + text_compare_audio_dir, dropdown_speaker_verification): global hide_voice_similarity_dir text_work_space_dir, text_base_audio_path, text_compare_audio_dir \ = common.batch_clean_paths([text_work_space_dir, text_base_audio_path, text_compare_audio_dir]) @@ -327,10 +330,12 @@ def similarity_audio_output(text_work_space_dir, text_role, text_base_audio_path raise Exception("基准音频路径不能为空") if text_compare_audio_dir is None or text_compare_audio_dir == '': raise Exception("待分析的音频所在目录不能为空") + if dropdown_speaker_verification is None or dropdown_speaker_verification == '': + raise Exception("说话人验证模型不能为空") time_consuming, (similarity_list, similarity_file, similarity_file_dir) \ - = time_util.time_monitor(start_similarity_analysis)(base_role_dir, - text_compare_audio_dir, text_base_audio_path, True) + = time_util.time_monitor(start_similarity_analysis)(base_role_dir,text_compare_audio_dir, + dropdown_speaker_verification, text_base_audio_path, True) if similarity_list is None: raise Exception("相似度分析失败") @@ -657,8 +662,14 @@ def init_ui(): scale=4) button_convert_from_list_result_dir = gr.Button(i18n("打开目录"), variant="primary", scale=1) gr.Markdown(value=i18n("1.2：选择基准音频，执行相似度匹配，并分段随机抽样")) - text_sample_dir = gr.Text(label=i18n("参考音频抽样目录"), value=init.text_sample_dir_default, - interactive=True) + with gr.Row(): + text_sample_dir = gr.Text(label=i18n("参考音频抽样目录"), value=init.text_sample_dir_default, + interactive=True) + dropdown_speaker_verification_1 = gr.Dropdown(label=i18n("说话人确认算法"), + choices=list( + model_manager.speaker_verification_models.keys()), + value='speech_campplus_sv_zh-cn_16k-common', + interactive=True) button_convert_from_list_result_dir.click(open_file, [text_sample_dir], []) button_convert_from_list.click(convert_from_list, [text_work_space_dir, text_role, text_list_input], [text_convert_from_list_info, text_sample_dir]) @@ -975,6 +986,11 @@ def init_ui(): with gr.Row(): text_base_audio_path = gr.Text(label=i18n("请输入基准音频"), value="") text_compare_audio_dir = gr.Text(label=i18n("请输入待比较的音频文件目录"), value="") + dropdown_speaker_verification_2 = gr.Dropdown(label=i18n("说话人确认算法"), + choices=list( + model_manager.speaker_verification_models.keys()), + value='speech_campplus_sv_zh-cn_16k-common', + interactive=True) with gr.Row(): button_similarity_audio_output = gr.Button(i18n("输出相似度-参考音频到临时目录"), variant="primary", scale=4) @@ -983,7 +999,7 @@ def init_ui(): button_similarity_audio_output_result_open = gr.Button(i18n("打开目录"), variant="primary", scale=1) button_similarity_audio_output.click(similarity_audio_output, [text_work_space_dir, text_role, text_base_audio_path, - text_compare_audio_dir], [text_similarity_audio_output_info]) + text_compare_audio_dir, dropdown_speaker_verification_2], [text_similarity_audio_output_info]) button_similarity_audio_output_result_open.click(lambda: open_file(hide_voice_similarity_dir), [], []) gr.Markdown(value=i18n("4.2：如果发现存在低音质的推理音频，那么就去参考音频目录下，把原参考音频删了")) gr.Markdown(value=i18n("4.3：删除参考音频之后，按下面的操作，会将推理音频目录下对应的音频也删掉")) @@ -1005,7 +1021,7 @@ def init_ui(): [text_work_space_dir, text_role, text_template, text_refer_audio_file_dir], [text_create_config_info]) button_create_config_result_open.click(lambda: open_file(hide_config_file), [], []) - button_sample.click(sample, [text_work_space_dir, text_role, text_sample_dir, text_base_voice_path, + button_sample.click(sample, [text_work_space_dir, text_role, text_sample_dir, dropdown_speaker_verification_1, text_base_voice_path, slider_subsection_num, slider_sample_num, checkbox_similarity_output], [text_sample_info, text_refer_audio_file_dir]) button_sample_result_open.click(open_file, [text_refer_audio_file_dir], []) diff --git a/Ref_Audio_Selector/tool/speaker_verification/voice_similarity.py b/Ref_Audio_Selector/tool/speaker_verification/voice_similarity.py index 3c7ec71..4f11fdf 100644 --- a/Ref_Audio_Selector/tool/speaker_verification/voice_similarity.py +++ b/Ref_Audio_Selector/tool/speaker_verification/voice_similarity.py @@ -6,18 +6,24 @@ import platform import Ref_Audio_Selector.config_param.config_params as params import Ref_Audio_Selector.config_param.log_config as log_config from Ref_Audio_Selector.common.time_util import timeit_decorator +from Ref_Audio_Selector.common.model_manager import speaker_verification_models as models from modelscope.pipelines import pipeline -sv_pipeline = pipeline( - task='speaker-verification', - model='Ref_Audio_Selector/tool/speaker_verification/models/speech_campplus_sv_zh-cn_16k-common', - model_revision='v1.0.0' -) + +def init_model(model_type='speech_campplus_sv_zh-cn_16k-common'): + log_config.logger.info(f'人声识别模型类型：{model_type}') + return pipeline( + task=models[model_type]['task'], + model=models[model_type]['model'], + model_revision=models[model_type]['model_revision'] + ) @timeit_decorator -def compare_audio_and_generate_report(reference_audio_path, comparison_dir_path, output_file_path): +def compare_audio_and_generate_report(reference_audio_path, comparison_dir_path, output_file_path, model_type): + sv_pipeline = init_model(model_type) + # Step 1: 获取比较音频目录下所有音频文件的路径 comparison_audio_paths = [os.path.join(comparison_dir_path, f) for f in os.listdir(comparison_dir_path) if f.endswith('.wav')] @@ -113,6 +119,10 @@ def parse_arguments(): parser.add_argument("-o", "--output_file", type=str, required=True, help="Path to the output file where results will be written.") + # Model Type + parser.add_argument("-m", "--model_type", type=str, required=True, + help="Path to the model type.") + return parser.parse_args() @@ -122,6 +132,7 @@ if __name__ == '__main__': reference_audio_path=cmd.reference_audio, comparison_dir_path=cmd.comparison_dir, output_file_path=cmd.output_file, + model_type=cmd.model_type, ) # compare_audio_and_generate_report(