Update app.py

app.py

@@ -2,82 +2,22 @@ import gradio as gr
 import numpy as np
 import torch
 import torchaudio
+import re
 from funasr import AutoModel
 
-print("正在加载SenseVoice-Small模型，请稍候...")
 model = AutoModel(
     model="iic/SenseVoiceSmall",
     vad_model="iic/speech_fsmn_vad_zh-cn-16k-common-pytorch",
     vad_kwargs={"max_single_segment_time": 30000},
     trust_remote_code=True,
 )
-print("模型加载成功！")
-
-emo_dict = {
-    "<|HAPPY|>": "😊",
-    "<|SAD|>": "😔",
-    "<|ANGRY|>": "😡",
-    "<|NEUTRAL|>": "",
-    "<|FEARFUL|>": "😰",
-    "<|DISGUSTED|>": "🤢",
-    "<|SURPRISED|>": "😮",
-}
-
-event_dict = {
-    "<|BGM|>": "🎼",
-    "<|Speech|>": "",
-    "<|Applause|>": "👏",
-    "<|Laughter|>": "😀",
-    "<|Cry|>": "😭",
-    "<|Sneeze|>": "🤧",
-    "<|Breath|>": "",
-    "<|Cough|>": "🤧",
-}
-
-emoji_dict = {
-    "<|nospeech|><|Event_UNK|>": "❓",
-    "<|zh|>": "",
-    "<|en|>": "",
-    "<|yue|>": "",
-    "<|ja|>": "",
-    "<|ko|>": "",
-    "<|nospeech|>": "",
-    "<|HAPPY|>": "😊",
-    "<|SAD|>": "😔",
-    "<|ANGRY|>": "😡",
-    "<|NEUTRAL|>": "",
-    "<|BGM|>": "🎼",
-    "<|Speech|>": "",
-    "<|Applause|>": "👏",
-    "<|Laughter|>": "😀",
-    "<|FEARFUL|>": "😰",
-    "<|DISGUSTED|>": "🤢",
-    "<|SURPRISED|>": "😮",
-    "<|Cry|>": "😭",
-    "<|EMO_UNKNOWN|>": "",
-    "<|Sneeze|>": "🤧",
-    "<|Breath|>": "",
-    "<|Cough|>": "😷",
-    "<|Sing|>": "",
-    "<|Speech_Noise|>": "",
-    "<|withitn|>": "",
-    "<|woitn|>": "",
-    "<|GBG|>": "",
-    "<|Event_UNK|>": "",
-}
-
-lang_dict = {
-    "<|zh|>": "<|lang|>",
-    "<|en|>": "<|lang|>",
-    "<|yue|>": "<|lang|>",
-    "<|ja|>": "<|lang|>",
-    "<|ko|>": "<|lang|>",
-    "<|nospeech|>": "<|lang|>",
-}
 
+emo_dict = {"<|HAPPY|>":"😊", "<|SAD|>":"😔", "<|ANGRY|>":"😡", "<|NEUTRAL|>":"", "<|FEARFUL|>":"😰", "<|DISGUSTED|>":"🤢", "<|SURPRISED|>":"😮"}
+event_dict = {"<|BGM|>":"🎼", "<|Speech|>":"", "<|Applause|>":"👏", "<|Laughter|>":"😀", "<|Cry|>":"😭", "<|Sneeze|>":"🤧", "<|Breath|>":"", "<|Cough|>":"🤧"}
+emoji_dict = {"<|nospeech|><|Event_UNK|>":"❓", "<|zh|>":"", "<|en|>":"", "<|yue|>":"", "<|ja|>":"", "<|ko|>":"", "<|nospeech|>":"", "<|HAPPY|>":"😊", "<|SAD|>":"😔", "<|ANGRY|>":"😡", "<|NEUTRAL|>":"", "<|BGM|>":"🎼", "<|Speech|>":"", "<|Applause|>":"👏", "<|Laughter|>":"😀", "<|FEARFUL|>":"😰", "<|DISGUSTED|>":"🤢", "<|SURPRISED|>":"😮", "<|Cry|>":"😭", "<|EMO_UNKNOWN|>":"", "<|Sneeze|>":"🤧", "<|Breath|>":"", "<|Cough|>":"😷", "<|Sing|>":"", "<|Speech_Noise|>":"", "<|withitn|>":"", "<|woitn|>":"", "<|GBG|>":"", "<|Event_UNK|>":""}
+lang_dict = {"<|zh|>":"<|lang|>", "<|en|>":"<|lang|>", "<|yue|>":"<|lang|>", "<|ja|>":"<|lang|>", "<|ko|>":"<|lang|>", "<|nospeech|>":"<|lang|>"}
 emo_set = {"😊", "😔", "😡", "😰", "🤢", "😮"}
-event_set = {"🎼", "👏", "😀", "😭", "🤧", "😷"}
-
+event_set = = {"🎼", "👏", "😀", "😭", "🤧", "😷"}
 
 def format_str_v2(s):
     sptk_dict = {}
@@ -86,115 +26,79 @@ def format_str_v2(s):
         s = s.replace(sptk, "")
     emo = "<|NEUTRAL|>"
     for e in emo_dict:
-        if sptk_dict.get(e, 0) > sptk_dict.get(emo, 0):
-            emo = e
+        if sptk_dict.get(e, 0) > sptk_dict.get(emo, 0): emo = e
     for e in event_dict:
-        if sptk_dict.get(e, 0) > 0:
-            s = event_dict[e] + s
+        if sptk_dict.get(e, 0) > 0: s = event_dict[e] + s
     s = s + emo_dict.get(emo, "")
-
     for emoji in emo_set.union(event_set):
-        s = s.replace(" " + emoji, emoji)
-        s = s.replace(emoji + " ", emoji)
+        s = s.replace(" " + emoji, emoji).replace(emoji + " ", emoji)
     return s.strip()
 
-
 def format_str_v3(s):
-    def get_emo(s):
-        return s[-1] if s and s[-1] in emo_set else None
-
-    def get_event(s):
-        return s[0] if s and s[0] in event_set else None
-
+    def get_emo(s): return s[-1] if s and s[-1] in emo_set else None
+    def get_event(s): return s[0] if s and s[0] in event_set else None
     s = s.replace("<|nospeech|><|Event_UNK|>", "❓")
-    for lang in lang_dict:
-        s = s.replace(lang, "<|lang|>")
+    for lang in lang_dict: s = s.replace(lang, "<|lang|>")
     s_list = [format_str_v2(s_i).strip(" ") for s_i in s.split("<|lang|>")]
     new_s = " " + (s_list[0] if s_list else "")
     cur_ent_event = get_event(new_s)
     for i in range(1, len(s_list)):
-        if not s_list[i]:
-            continue
-        if get_event(s_list[i]) == cur_ent_event and cur_ent_event is not None:
-            s_list[i] = s_list[i][1:]
+        if not s_list[i]: continue
+        if get_event(s_list[i]) == cur_ent_event and cur_ent_event is not None: s_list[i] = s_list[i][1:]
         cur_ent_event = get_event(s_list[i])
-        if get_emo(s_list[i]) is not None and get_emo(s_list[i]) == get_emo(new_s):
-            new_s = new_s[:-1]
+        if get_emo(s_list[i]) is not None and get_emo(s_list[i]) == get_emo(new_s): new_s = new_s[:-1]
         new_s += s_list[i].strip().lstrip()
-    new_s = new_s.replace("The.", " ")
     return new_s.strip()
 
-
-def model_inference(audio_input, language):
-    if audio_input is None:
-        return "错误：请先上传或录制音频。"
-
+def model_inference(audio_input, language, output_format, use_itn, merge_vad, merge_length, ban_emo_unk):
+    if audio_input is None: return "错误：请上传或录制音频。"
     fs, input_wav = audio_input
-    input_wav = input_wav.astype(np.float32) / 32768.0
-
-    if len(input_wav.shape) > 1:
-        input_wav = input_wav.mean(-1)
-
+    input_wav = input_wav.astype(np.float32) / np.iinfo(np.int16).max
+    if len(input_wav.shape) > 1: input_wav = input_wav.mean(-1)
     if fs != 16000:
-        print(f"音频采样率不是16k，正在重采样: {fs}")
         resampler = torchaudio.transforms.Resample(orig_freq=fs, new_freq=16000)
         input_wav = resampler(torch.from_numpy(input_wav)).numpy()
-
-    print(f"开始识别... 语言: {language}")
-
-    res = model.generate(
-        input=input_wav,
-        cache={},
-        language=language,
-        use_itn=True,
-        batch_size_s=60,
-        merge_vad=True,
-    )
-
+    res = model.generate(input=input_wav, cache={}, language=language, use_itn=use_itn, batch_size_s=60, merge_vad=merge_vad, merge_length_s=merge_length, ban_emo_unk=ban_emo_unk)
     raw_text = res[0].get("text", "未能识别出文本。")
-    print(f"原始输出: {raw_text}")
-
-    # 使用格式化函数处理文本
-    formatted_text = format_str_v3(raw_text)
-    print(f"格式化输出: {formatted_text}")
-
-    return formatted_text
-
-
-html_content = """
-<div style="text-align: center;">
-    <h2 style="font-size: 22px;">语音理解模型： SenseVoice-Small</h2>
-    <p style="font-size: 16px;">
-        这是一个部署在Hugging Face上的SenseVoice-Small语音识别模型。<br>
-        它支持中、英、粤、日、韩等多语言识别，并能识别情感和声音事件。
-    </p>
-    <p style="font-size: 16px;">
-        <b>用法:</b> 上传音频或使用麦克风录音，在下方“配置”中选择识别语言，然后点击“开始”。
-    </p>
+    if output_format == "原始富文本": return raw_text
+    elif output_format == "纯净文本": return ' '.join(re.sub(r'<\|[^>]*>', '', raw_text).split())
+    else: return format_str_v3(raw_text)
+
+html_intro = """
+<div>
+    <h2 style="font-size: 22px;margin-left: 0px;">语音理解模型： SenseVoice-Small</h2>
+    <p style="font-size: 18px;margin-left: 20px;">
+SenseVoice-Small 是一种仅包含编码器的语音基础模型，专为快速语音理解而设计。它包含多种功能，包括自动语音识别 (ASR)、口语识别 (LID)、语音情感识别 (SER) 和声学事件检测 (AED)。 SenseVoice-Small支持中文、英语、粤语、日语、韩语等多语言识别。此外，它还提供极低的推理延迟，执行速度比 Whisper-small 快 7 倍，比 Whisper-large 快 17 倍。</p>
+    <h2 style="font-size: 22px;margin-left: 0px;">用法</h2> <p style="font-size: 18px;margin-left: 20px;">上传音频文件或通过麦克风输入，然后选择任务和语言。音频被转录成相应的文本以及相关的情绪（😊快乐，😡愤怒/兴奋，😔悲伤）和声音事件类型（😀笑声，🎼音乐，👏掌声，🤧咳嗽和打喷嚏，😔哭泣）。事件标签放在文本的前面，情感放在文本的后面。</p>
+	<h2 style="font-size: 22px;margin-left: 0px;">目录</h2>
+	<p style="font-size: 18px;margin-left: 20px;"><a href="https://github.com/FunAudioLLM/SenseVoice" target="_blank">SenseVoice</a>：多语言语音理解模型</p>
+	<p style="font-size: 18px;margin-left: 20px;"><a href="https://github.com/modelscope/FunASR" target="_blank">FunASR</a>：基本语音识别工具包</p>
+	<p style="font-size: 18px;margin-left: 20px;"><a href="https://github.com/FunAudioLLM/CosyVoice" target="_blank">CosyVoice</a>：高质量多语言TTS模型</p>
+</div>
+<div style="text-align: center; font-size: 15px; font-weight: bold; color: red; margin-bottom: 20px;">
+        ⚠️ 该演示仅供学术研究和体验使用。
 </div>
 """
 
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.HTML(html_content)
+lang_options = [("自动检测", "auto"), ("中文", "zh"), ("English", "en"), ("粤语", "yue"), ("日本語", "ja"), ("한국어", "ko"), ("无语音", "nospeech")]
+example_list = [[f"{model.model_path}/example/zh.mp3", "auto"], [f"{model.model_path}/example/en.mp3", "auto"], [f"{model.model_path}/example/yue.mp3", "auto"], [f"{model.model_path}/example/ja.mp3", "auto"], [f"{model.model_path}/example/ko.mp3", "auto"], [f"{model.model_path}/example/rich_1.wav", "auto"], [f"{model.model_path}/example/rich_2.wav", "auto"], [f"{model.model_path}/example/longwav_1.wav", "auto"]]
 
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.HTML(html_intro)
     with gr.Row():
-        with gr.Column():
-            audio_inputs = gr.Audio(label="上传音频或使用麦克风录音")
-
-            with gr.Accordion("配置", open=True):
-                language_inputs = gr.Dropdown(
-                    choices=["auto", "zh", "en", "yue", "ja", "ko", "nospeech"],
-                    value="auto",
-                    label="选择语言 (Select Language)",
-                )
-
+        with gr.Column(scale=2):
+            audio_inputs = gr.Audio(label="上传音频或使用麦克风")
+            with gr.Accordion("高级设置", open=False):
+                output_format_dropdown = gr.Dropdown(choices=["Emoji格式化", "纯净文本", "原始富文本"], value="Emoji格式化", label="选择输出格式")
+                language_inputs = gr.Dropdown(choices=lang_options, value="auto", label="识别语言")
+                use_itn_checkbox = gr.Checkbox(value=True, label="启用标点和逆文本正规化")
+                merge_vad_checkbox = gr.Checkbox(value=True, label="合并VAD切分的音频片段")
+                merge_length_slider = gr.Slider(minimum=5, maximum=30, value=15, step=1, label="合并后片段最大长度(秒)")
+                ban_emo_unk_checkbox = gr.Checkbox(value=False, label="禁用未知情感标签")
             fn_button = gr.Button("开始识别", variant="primary")
+        with gr.Column(scale=3):
+            text_outputs = gr.Textbox(label="识别结果", lines=20)
+            gr.Examples(examples=example_list, inputs=[audio_inputs, language_inputs], label="示例音频 (点击即可试用)")
+    fn_button.click(fn=model_inference, inputs=[audio_inputs, language_inputs, output_format_dropdown, use_itn_checkbox, merge_vad_checkbox, merge_length_slider, ban_emo_unk_checkbox], outputs=text_outputs)
 
-    with gr.Row():
-        text_outputs = gr.Textbox(label="识别结果", lines=20)
-
-    fn_button.click(
-        fn=model_inference, inputs=[audio_inputs, language_inputs], outputs=text_outputs
-    )
-
-demo.launch()
+demo.launch()