Update app.py

app.py

@@ -1,33 +1,3 @@
-# import gradio as gr
-# import torch
-# from transformers import pipeline, AutoTokenizer
-# from nemo.collections.asr.models import EncDecMultiTaskModel
-
-# # load model
-# canary_model = EncDecMultiTaskModel.from_pretrained('nvidia/canary-1b')
-
-# # update dcode params
-# decode_cfg = canary_model.cfg.decoding
-# decode_cfg.beam.beam_size = 1
-# canary_model.change_decoding_strategy(decode_cfg)
-
-# pipe = pipeline(
-#     "automatic-speech-recognition", 
-#     model="nvidia/canary-1b"
-# )
-
-# # pipe = pipeline(
-# #     "text-generation", 
-# #     model="QuantFactory/Meta-Llama-3-8B-Instruct-GGUF", 
-# #     model_kwargs={"torch_dtype": torch.bfloat16}, 
-# #     device_map="auto"
-# # )
-
-# gr.Interface.from_pipeline(pipe,
-#                            title="ASR",
-#                            description="Using pipeline with Canary-1B",
-#                            ).launch(inbrowser=True)
-
 import gradio as gr
 import json
 import librosa
@@ -171,6 +141,104 @@ def transcribe(audio_filepath, src_lang, tgt_lang, pnc):
 
 	return output_text
 
+# add logic to make sure dropdown menus only suggest valid combos
+def on_src_or_tgt_lang_change(src_lang_value, tgt_lang_value, pnc_value):
+	"""Callback function for when src_lang or tgt_lang dropdown menus are changed.
+
+	Args:
+		src_lang_value(string), tgt_lang_value (string), pnc_value(bool) - the current 
+			chosen "values" of each Gradio component
+	Returns:
+		src_lang, tgt_lang, pnc - these are the new Gradio components that will be displayed
+	
+	Note: I found the required logic is easier to understand if you think about the possible src & tgt langs as
+	a matrix, e.g. with English, Spanish, French, German as the langs, and only transcription in the same language,
+	and X -> English and English -> X translation being allowed, the matrix looks like the diagram below ("Y" means it is
+	allowed to go into that state).
+	It is easier to understand the code if you think about which state you are in, given the current src_lang_value and
+	tgt_lang_value, and then which states you can go to from there.
+
+			tgt lang
+			- |EN |ES |FR |DE
+			------------------
+			EN| Y | Y | Y | Y
+			------------------
+		src 	ES| Y | Y |   |
+		lang	------------------
+			FR| Y |   | Y |
+			------------------
+			DE| Y |   |   | Y
+	"""
+
+	if src_lang_value == "English" and tgt_lang_value == "English":
+		# src_lang and tgt_lang can go anywhere
+		src_lang = gr.Dropdown(
+			choices=["English", "Spanish", "French", "German"],
+			value=src_lang_value,
+			label="Input audio is spoken in:"
+		)
+		tgt_lang = gr.Dropdown(
+			choices=["English", "Spanish", "French", "German"],
+			value=tgt_lang_value,
+			label="Transcribe in language:"
+		)
+	elif src_lang_value == "English": 
+		# src is English & tgt is non-English
+		# => src can only be English or current tgt_lang_values
+		# & tgt can be anything
+		src_lang = gr.Dropdown(
+			choices=["English", tgt_lang_value],
+			value=src_lang_value,
+			label="Input audio is spoken in:"
+		)
+		tgt_lang = gr.Dropdown(
+			choices=["English", "Spanish", "French", "German"],
+			value=tgt_lang_value,
+			label="Transcribe in language:"
+		)
+	elif tgt_lang_value == "English": 
+		# src is non-English & tgt is English
+		# => src can be anything
+		# & tgt can only be English or current src_lang_value
+		src_lang = gr.Dropdown(
+			choices=["English", "Spanish", "French", "German"],
+			value=src_lang_value,
+			label="Input audio is spoken in:"
+		)
+		tgt_lang = gr.Dropdown(
+			choices=["English", src_lang_value],
+			value=tgt_lang_value,
+			label="Transcribe in language:"
+		)
+	else:
+		# both src and tgt are non-English
+		# => both src and tgt can only be switch to English or themselves
+		src_lang = gr.Dropdown(
+			choices=["English", src_lang_value],
+			value=src_lang_value,
+			label="Input audio is spoken in:"
+		)
+		tgt_lang = gr.Dropdown(
+			choices=["English", tgt_lang_value],
+			value=tgt_lang_value,
+			label="Transcribe in language:"
+		)
+	# let pnc be anything if src_lang_value == tgt_lang_value, else fix to True
+	if src_lang_value == tgt_lang_value:
+		pnc = gr.Checkbox(
+			value=pnc_value,
+			label="Punctuation & Capitalization in transcript?",
+			interactive=True
+		)
+	else:
+		pnc = gr.Checkbox(
+			value=True,
+			label="Punctuation & Capitalization in transcript?",
+			interactive=False
+		)
+	return src_lang, tgt_lang, pnc
+
+
 with gr.Blocks(
 	title="NeMo Canary Model",
 	css="""
@@ -230,32 +298,32 @@ with gr.Blocks(
 				elem_id="model_output_text_box",
 			)
 
-        with gr.Row():
-
-    		gr.HTML(
-    			"<p style='text-align: center'>"
-    				"🐤 <a href='https://huggingface.co/nvidia/canary-1b' target='_blank'>Canary model</a> | "
-    				"🧑‍💻 <a href='https://github.com/NVIDIA/NeMo' target='_blank'>NeMo Repository</a>"
-    			"</p>"
-    		)
-
-    	go_button.click(
-    		fn=transcribe, 
-    		inputs = [audio_file, src_lang, tgt_lang, pnc],
-    		outputs = [model_output_text_box]
-    	)
-    
-    	# call on_src_or_tgt_lang_change whenever src_lang or tgt_lang dropdown menus are changed	
-    	src_lang.change(
-    		fn=on_src_or_tgt_lang_change,
-    		inputs=[src_lang, tgt_lang, pnc],
-    		outputs=[src_lang, tgt_lang, pnc],
-    	)
-    	tgt_lang.change(
-    		fn=on_src_or_tgt_lang_change,
-    		inputs=[src_lang, tgt_lang, pnc],
-    		outputs=[src_lang, tgt_lang, pnc],
-    	)
+	with gr.Row():
+
+		gr.HTML(
+			"<p style='text-align: center'>"
+				"🐤 <a href='https://huggingface.co/nvidia/canary-1b' target='_blank'>Canary model</a> | "
+				"🧑‍💻 <a href='https://github.com/NVIDIA/NeMo' target='_blank'>NeMo Repository</a>"
+			"</p>"
+		)
+
+	go_button.click(
+		fn=transcribe, 
+		inputs = [audio_file, src_lang, tgt_lang, pnc],
+		outputs = [model_output_text_box]
+	)
+
+	# call on_src_or_tgt_lang_change whenever src_lang or tgt_lang dropdown menus are changed	
+	src_lang.change(
+		fn=on_src_or_tgt_lang_change,
+		inputs=[src_lang, tgt_lang, pnc],
+		outputs=[src_lang, tgt_lang, pnc],
+	)
+	tgt_lang.change(
+		fn=on_src_or_tgt_lang_change,
+		inputs=[src_lang, tgt_lang, pnc],
+		outputs=[src_lang, tgt_lang, pnc],
+	)
 
 
 demo.queue()