Update app.py

app.py

@@ -32,7 +32,6 @@ models_en = {
     name: onnx_asr.load_model(name)
     for name in [
         "nemo-parakeet-ctc-0.6b",
-        "nemo-parakeet-rnnt-0.6b",
         "nemo-parakeet-tdt-0.6b-v2",
     ]
 }
@@ -45,7 +44,7 @@ def recognize(audio: tuple[int, np.ndarray], models, language):
         return None
 
     sample_rate, waveform = audio
-    length = waveform.shape[-1] / sample_rate
+    length = waveform.shape[0] / sample_rate
     logger.debug("recognize: length %.3f, sample_rate %s, waveform.shape %s.", length, sample_rate, waveform.shape)
     try:
         waveform = waveform.astype(np.float32) / 2 ** (8 * waveform.itemsize - 1)
@@ -60,7 +59,7 @@ def recognize(audio: tuple[int, np.ndarray], models, language):
             result = model.recognize(waveform, sample_rate=sample_rate, language=language)
             time = timer() - start
             logger.debug("recognized by %s: result '%s', time %.3f s.", name, result, time)
-            results.append([name, result, f"{time:.3f} s."])
+            results.append([name, result])
 
     except Exception as e:
         raise gr.Error(f"{e} Audio: sample_rate: {sample_rate}, waveform.shape: {waveform.shape}.") from e
@@ -81,7 +80,8 @@ def recognize_with_vad(audio: tuple[int, np.ndarray], name: str):
         return None
 
     sample_rate, waveform = audio
-    logger.debug("recognize: sample_rate %s, waveform.shape %s.", sample_rate, waveform.shape)
+    length = waveform.shape[0] / sample_rate
+    logger.debug("recognize: length %.3f, sample_rate %s, waveform.shape %s.", length, sample_rate, waveform.shape)
     try:
         waveform = waveform.astype(np.float32) / 2 ** (8 * waveform.itemsize - 1)
         if waveform.ndim == 2:
@@ -104,14 +104,14 @@ with gr.Blocks() as recognize_short:
         gr.ClearButton(audio)
         btn_ru = gr.Button("Recognize (ru)", variant="primary")
         btn_en = gr.Button("Recognize (en)", variant="primary")
-    output = gr.Dataframe(headers=["model", "result", "time"], wrap=True)
+    output = gr.Dataframe(headers=["model", "result"], wrap=True)
     btn_ru.click(fn=recognize_ru, inputs=audio, outputs=output)
     btn_en.click(fn=recognize_en, inputs=audio, outputs=output)
 
 
 with gr.Blocks() as recognize_long:
+    gr.Markdown("For better results, you need to adjust the VAD parameters.")
     name = gr.Dropdown(models_vad.keys(), label="Model")
-    # lang = gr.Label()
     audio = gr.Audio(min_length=1, max_length=300)
     with gr.Row():
         gr.ClearButton(audio)
@@ -130,7 +130,7 @@ with gr.Blocks() as recognize_long:
 
     name.change(on_model_change, inputs=name, outputs=audio)
 
-with gr.Blocks() as demo:
+with gr.Blocks(title="onnx-asr demo") as demo:
     gr.Markdown("""
     # ASR demo using onnx-asr
     **[onnx-asr](https://github.com/istupakov/onnx-asr)** is a Python package for Automatic Speech Recognition using ONNX models.
@@ -155,7 +155,6 @@ with gr.Blocks() as demo:
         * `alphacep/vosk-model-small-ru` - Alpha Cephei Vosk 0.52-small-ru ([origin](https://huggingface.co/alphacep/vosk-model-small-ru))
         ## English ASR models
         * `nemo-parakeet-ctc-0.6b` - Nvidia Parakeet CTC 0.6B (en) ([origin](https://huggingface.co/nvidia/parakeet-ctc-0.6b), [onnx](https://huggingface.co/istupakov/parakeet-ctc-0.6b-onnx))
-        * `nemo-parakeet-rnnt-0.6b` - Nvidia Parakeet RNNT 0.6B (en) ([origin](https://huggingface.co/nvidia/parakeet-rnnt-0.6b), [onnx](https://huggingface.co/istupakov/parakeet-rnnt-0.6b-onnx))
         * `nemo-parakeet-tdt-0.6b-v2` - Nvidia Parakeet TDT 0.6B V2 (en) ([origin](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2), [onnx](https://huggingface.co/istupakov/parakeet-tdt-0.6b-v2-onnx))
         * `whisper-base` - OpenAI Whisper Base exported with onnxruntime ([origin](https://huggingface.co/openai/whisper-base), [onnx](https://huggingface.co/istupakov/whisper-base-onnx))
         ## VAD models