Update app.py

app.py

@@ -14,13 +14,13 @@ logger.info("onnx_asr version: %s", version("onnx_asr"))
 
 vad = onnx_asr.load_vad("silero")
 
-models_multilang = {name: onnx_asr.load_model(name) for name in ["whisper-base", "nemo-parakeet-tdt-0.6b-v3"]}
+models_multilang = {name: onnx_asr.load_model(name) for name in ["whisper-base"]} | {
+    name: onnx_asr.load_model(name, quantization="int8") for name in ["nemo-parakeet-tdt-0.6b-v3", "nemo-canary-1b-v2"]
+}
 
 models_ru = {
     name: onnx_asr.load_model(name)
     for name in [
-        "gigaam-v2-ctc",
-        "gigaam-v2-rnnt",
         "gigaam-v3-ctc",
         "gigaam-v3-rnnt",
         "gigaam-v3-e2e-ctc",
@@ -29,11 +29,12 @@ models_ru = {
         "nemo-fastconformer-ru-rnnt",
         "alphacep/vosk-model-ru",
         "alphacep/vosk-model-small-ru",
+        "t-tech/t-one",
     ]
 }
 
 models_en = {
-    name: onnx_asr.load_model(name)
+    name: onnx_asr.load_model(name, quantization="int8")
     for name in [
         "nemo-parakeet-tdt-0.6b-v2",
     ]
@@ -42,10 +43,14 @@ models_en = {
 models_vad = models_multilang | models_ru | models_en
 
 
-def recognize(audio: tuple[int, np.ndarray], models, language):
+def recognize(audio: tuple[int, np.ndarray], models, language: str):
     if audio is None:
         return None
 
+    valid_res = gr.validators.is_audio_correct_length(audio, min_length=1, max_length=10)
+    if not valid_res["is_valid"]:
+        raise gr.Error(valid_res["message"])
+
     sample_rate, waveform = audio
     length = waveform.shape[0] / sample_rate
     logger.debug("recognize: length %.3f, sample_rate %s, waveform.shape %s.", length, sample_rate, waveform.shape)
@@ -59,6 +64,7 @@ def recognize(audio: tuple[int, np.ndarray], models, language):
             if length > 20 and name == "alphacep/vosk-model-small-ru":
                 gr.Warning(f"Model {name} only supports audio no longer than 20 s.")
                 continue
+
             start = timer()
             result = model.recognize(waveform, sample_rate=sample_rate, language=language)
             time = timer() - start
@@ -83,6 +89,10 @@ def recognize_with_vad(audio: tuple[int, np.ndarray], name: str):
     if audio is None:
         return None
 
+    valid_res = gr.validators.is_audio_correct_length(audio, min_length=1, max_length=300)
+    if not valid_res["is_valid"]:
+        raise gr.Error(valid_res["message"])
+
     sample_rate, waveform = audio
     length = waveform.shape[0] / sample_rate
     logger.debug("recognize: length %.3f, sample_rate %s, waveform.shape %s.", length, sample_rate, waveform.shape)
@@ -103,9 +113,8 @@ def recognize_with_vad(audio: tuple[int, np.ndarray], name: str):
 
 
 with gr.Blocks() as recognize_short:
-    audio = gr.Audio(min_length=1, max_length=30)
+    audio = gr.Audio()
     with gr.Row():
-        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"], wrap=True)
@@ -116,9 +125,8 @@ with gr.Blocks() as recognize_short:
 with gr.Blocks() as recognize_long:
     gr.Markdown("The default VAD parameters are used. For best results, you should adjust the VAD parameters in your app.")
     name = gr.Dropdown(models_vad.keys(), label="Model")
-    audio = gr.Audio(min_length=1, max_length=300)
+    audio = gr.Audio()
     with gr.Row():
-        gr.ClearButton(audio)
         btn = gr.Button("Recognize", variant="primary")
     output = gr.TextArea(label="result")
     btn.click(fn=recognize_with_vad, inputs=[audio, name], outputs=output)
@@ -130,7 +138,7 @@ with gr.Blocks() as recognize_long:
             label = f"Model {name} support only English language"
         else:
             label = None
-        return gr.Audio(min_length=1, max_length=300, label=label)
+        return gr.Audio(label=label)
 
     name.change(on_model_change, inputs=name, outputs=audio)
 
@@ -138,9 +146,10 @@ 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.
-    The package is written in pure Python with minimal dependencies (no `pytorch` or `transformers`).
-    Supports Parakeet TDT 0.6B V2 (En), Parakeet TDT 0.6B V3 (Multilingual) and GigaAM v2/v3 (Ru) models
-    (and many other modern [models](https://github.com/istupakov/onnx-asr?tab=readme-ov-file#supported-model-names)).
+    It's written in pure Python with minimal dependencies (no PyTorch, Transformers, or FFmpeg required).
+
+    Supports **Parakeet v2 (En) / v3 (Multilingual)**, **Canary v2 (Multilingual)** and **GigaAM v2/v3 (Ru)** models
+    (and many other modern [models](https://github.com/istupakov/onnx-asr?tab=readme-ov-file#supported-model-names)).   
     You can also use it with your own model if it has a supported architecture.
     """)
     gr.TabbedInterface(
@@ -153,21 +162,22 @@ with gr.Blocks(title="onnx-asr demo") as demo:
     with gr.Accordion("Models used in this demo:", open=False):
         gr.Markdown("""
         ## Russian ASR models
-        * `gigaam-v2-ctc` - Sber GigaAM v2 CTC ([origin](https://github.com/salute-developers/GigaAM), [onnx](https://huggingface.co/istupakov/gigaam-v2-onnx))
-        * `gigaam-v2-rnnt` - Sber GigaAM v2 RNN-T ([origin](https://github.com/salute-developers/GigaAM), [onnx](https://huggingface.co/istupakov/gigaam-v2-onnx))
         * `gigaam-v3-ctc` - Sber GigaAM v3 CTC ([origin](https://huggingface.co/ai-sage/GigaAM-v3), [onnx](https://huggingface.co/istupakov/gigaam-v3-onnx))
         * `gigaam-v3-rnnt` - Sber GigaAM v3 RNN-T ([origin](https://huggingface.co/ai-sage/GigaAM-v3), [onnx](https://huggingface.co/istupakov/gigaam-v3-onnx))
         * `gigaam-v3-e2e-ctc` - Sber GigaAM v3 E2E CTC ([origin](https://huggingface.co/ai-sage/GigaAM-v3), [onnx](https://huggingface.co/istupakov/gigaam-v3-onnx))
         * `gigaam-v3-e2e-rnnt` - Sber GigaAM v3 E2E RNN-T ([origin](https://huggingface.co/ai-sage/GigaAM-v3), [onnx](https://huggingface.co/istupakov/gigaam-v3-onnx))
         * `nemo-fastconformer-ru-ctc` - Nvidia FastConformer-Hybrid Large (ru) with CTC decoder ([origin](https://huggingface.co/nvidia/stt_ru_fastconformer_hybrid_large_pc), [onnx](https://huggingface.co/istupakov/stt_ru_fastconformer_hybrid_large_pc_onnx))
         * `nemo-fastconformer-ru-rnnt` - Nvidia FastConformer-Hybrid Large (ru) with RNN-T decoder ([origin](https://huggingface.co/nvidia/stt_ru_fastconformer_hybrid_large_pc), [onnx](https://huggingface.co/istupakov/stt_ru_fastconformer_hybrid_large_pc_onnx))
-        * `nemo-parakeet-tdt-0.6b-v3` - Nvidia Parakeet TDT 0.6B V3 (multilingual) ([origin](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3), [onnx](https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx))
+        * `nemo-parakeet-tdt-0.6b-v3` - Nvidia Parakeet TDT 0.6B v3 (multilingual) ([origin](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3), [onnx](https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx))
+        * `nemo-canary-1b-v2` - Nvidia Canary 1B v2 (multilingual) ([origin](https://huggingface.co/nvidia/canary-1b-v2), [onnx](https://huggingface.co/istupakov/canary-1b-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))
         * `alphacep/vosk-model-ru` - Alpha Cephei Vosk 0.54-ru ([origin](https://huggingface.co/alphacep/vosk-model-ru))
         * `alphacep/vosk-model-small-ru` - Alpha Cephei Vosk 0.52-small-ru ([origin](https://huggingface.co/alphacep/vosk-model-small-ru))
+        * `t-tech/t-one` - T-Tech T-one ([origin](https://huggingface.co/t-tech/T-one))
         ## English ASR models
-        * `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))
-        * `nemo-parakeet-tdt-0.6b-v3` - Nvidia Parakeet TDT 0.6B V3 (multilingual) ([origin](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3), [onnx](https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-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))
+        * `nemo-parakeet-tdt-0.6b-v3` - Nvidia Parakeet TDT 0.6B v3 (multilingual) ([origin](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3), [onnx](https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx))
+        * `nemo-canary-1b-v2` - Nvidia Canary 1B v2 (multilingual) ([origin](https://huggingface.co/nvidia/canary-1b-v2), [onnx](https://huggingface.co/istupakov/canary-1b-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
         * `silero` - Silero VAD ([origin](https://github.com/snakers4/silero-vad), [onnx](https://huggingface.co/onnx-community/silero-vad))