Update consult.py

consult.py

@@ -36,6 +36,132 @@ os.environ["no_proxy"] = "localhost,127.0.0.1,::1"
 client = Client("Liusuthu/TextDepression")
 
 
+classifier = foreign_class(
+    source="pretrained_models/local-speechbrain/emotion-recognition-wav2vec2-IEMOCAP",  # ".\\emotion-recognition-wav2vec2-IEMOCAP"
+    pymodule_file="custom_interface.py",
+    classname="CustomEncoderWav2vec2Classifier",
+    savedir="pretrained_models/local-speechbrain/emotion-recognition-wav2vec2-IEMOCAP",
+)
+ASR_model = ParaformerOffline()
+vad = FSMNVad()
+punc = CttPunctuator()
+
+def text_api(text:str):
+    result = client.predict(
+        text,  # str  in '输入文字' Textbox component
+        api_name="/predict",
+    )
+    return result
+
+
+def get_text_score(text):
+    string=text_api(text)
+    part1 = str.partition(string, r"text")
+    want1 = part1[2]
+    label = want1[4:6]
+    part2 = str.partition(string, r"probability")
+    want2 = part2[2]
+    prob = float(want2[3:-4])
+    return label, prob
+
+def classify_continuous(audio):
+    print(type(audio))
+    print(audio)
+    sample_rate, signal = audio  # 这是语音的输入
+    signal = signal.astype(np.float32)
+    signal /= np.max(np.abs(signal))
+    sf.write("data/a.wav", signal, sample_rate)
+    signal, sample_rate = torchaudio.load("data/a.wav")
+    signal1 = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(
+        signal
+    )
+    torchaudio.save("data/out.wav", signal1, 16000, encoding="PCM_S", bits_per_sample=16)
+    Audio = "data/out.wav"
+    speech, sample_rate = AudioReader.read_wav_file(Audio)
+    if signal == "none":
+        return "none", "none", "haha"
+    else:
+        segments = vad.segments_offline(speech)
+        text_results = ""
+        for part in segments:
+            _result = ASR_model.infer_offline(
+                speech[part[0] * 16 : part[1] * 16], hot_words="任意热词 空格分开"
+            )
+            text_results += punc.punctuate(_result)[0]
+
+        out_prob, score, index, text_lab = classifier.classify_batch(signal1)
+        print(type(out_prob.squeeze(0).numpy()))
+        print(out_prob.squeeze(0).numpy())
+        print(type(text_lab[-1]))
+        print(text_lab[-1])
+        return text_results, out_prob.squeeze(0).numpy(), text_lab[-1], Audio
+
+
+
+#######################################################################
+#第四题专用函数：
+def text_score4(text):
+    string=text_api(text)
+    part1 = str.partition(string, r"text")
+    want1 = part1[2]
+    label = want1[4:6]
+    part2 = str.partition(string, r"probability")
+    want2 = part2[2]
+    prob = float(want2[3:-4])
+    if label=="正向":
+        score=-prob*10
+    else:
+        score=prob*10
+    return text,score,gr.Column(visible=True)
+
+
+def speech_score4(audio):
+    print(type(audio))
+    print(audio)
+    sample_rate, signal = audio  # 这是语音的输入
+    signal = signal.astype(np.float32)
+    signal /= np.max(np.abs(signal))
+    sf.write("data/a.wav", signal, sample_rate)
+    signal, sample_rate = torchaudio.load("data/a.wav")
+    signal1 = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(
+        signal
+    )
+    torchaudio.save("data/out.wav", signal1, 16000, encoding="PCM_S", bits_per_sample=16)
+    Audio = "data/out.wav"
+    speech, sample_rate = AudioReader.read_wav_file(Audio)
+    if signal == "none":
+        return "none", "none", "haha"
+    else:
+        segments = vad.segments_offline(speech)
+        text_results = ""
+        for part in segments:
+            _result = ASR_model.infer_offline(
+                speech[part[0] * 16 : part[1] * 16], hot_words="任意热词 空格分开"
+            )
+            text_results += punc.punctuate(_result)[0]
+
+        out_prob, score, index, text_lab = classifier.classify_batch(signal1)
+        print(type(out_prob.squeeze(0).numpy()))
+        print(out_prob.squeeze(0).numpy())
+        print(type(text_lab[-1]))
+        print(text_lab[-1])
+        #return text_results, out_prob.squeeze(0).numpy(), text_lab[-1], Audio
+        prob=out_prob.squeeze(0).numpy()
+        print(prob)
+        score2=10*prob[0]-10*prob[1]
+        print("score2",score2)
+        print(text_lab[-1])
+        text,score1=text_score4(text_results)
+        # text_emo=str(get_text_score(text_results))
+        print(text,score1)
+        score=score1+score2
+
+        return text,score,gr.Column(visible=True)
+
+
+
+#####################################################################
+
 # constants
 schema = "情感倾向[正向，负向]"  # Define the schema for sentence-level sentiment classification
 # 定义音频模态下权重