added dropdown to select which whisper model to use for transcription, added note that FFMPEG is required

README.md

@@ -223,7 +223,17 @@ To import a voice, click `Import Voice`. Remember to click `Refresh Voice List`
 
 This tab will contain a collection of sub-tabs pertaining to training.
 
-#### Configuration
+#### Prepare Dataset
+
+This section will aid in preparing the dataset for fine-tuning.
+
+With it, you simply select a voice, then click the button, and wait for the console to tell you it's done. The results will be saved to `./training/{voice name}/`.
+
+The web UI will leverage [openai/whisper](https://github.com/openai/whisper) to transcribe a given sample source, and split them into convenient pieces.
+
+**!**NOTE**!**: transcription leverages FFMPEG, so please make sure you either have an FFMPEG installed visible to your PATH, or drop the binary in the `./bin/` folder.
+
+#### Generate Configuration
 
 This will generate the YAML necessary to feed into training. For now, you can set:
 * `Batch Size`: size of batches for training, more batches = faster training, at the cost of higher VRAM. setting this to 1 will lead to problems
@@ -255,6 +265,10 @@ Below are settings that override the default launch arguments. Some of these req
 * `Embed Output Metadata`: enables embedding the settings and latents used to generate that audio clip inside that audio clip. Metadata is stored as a JSON string in the `lyrics` tag.
 * `Slimmer Computed Latents`: falls back to the original, 12.9KiB way of storing latents (without the extra bits required for using the CVVP model).
 * `Voice Fixer`: runs each generated audio clip through `voicefixer`, if available and installed.
+* `Use CUDA for Voice Fixer`: allows voicefixer to use CUDA. Speeds up cleaning the output, but at the cost of more VRAM consumed. Disable if you OOM.
+* `Device Override`: overrides the device name used to pass to PyTorch for hardware acceleration. You can use the accompanied `list_devices.py` script to map valid strings to GPU names. You can also pass `cpu` if you want to fallback to software mode.
+* `Whisper Model`: the specific model to use for Whisper transcription, when preparing a dataset to finetune with.
+
 * `Voice Latent Max Chunk Size`: during the voice latents calculation pass, this limits how large, in bytes, a chunk can be. Large values can run into VRAM OOM errors.
 * `Sample Batch Size`: sets the batch size when generating autoregressive samples. Bigger batches result in faster compute, at the cost of increased VRAM consumption. Leave to 0 to calculate a "best" fit.
 * `Concurrency Count`: how many Gradio events the queue can process at once. Leave this over 1 if you want to modify settings in the UI that updates other settings while generating audio clips.

src/utils.py

@@ -57,6 +57,7 @@ def setup_args():
 		'voice-fixer-use-cuda': True,
 		'force-cpu-for-conditioning-latents': False,
 		'device-override': None,
+		'whisper-model': "base",
 		'concurrency-count': 2,
 		'output-sample-rate': 44100,
 		'output-volume': 1,
@@ -80,6 +81,7 @@ def setup_args():
 	parser.add_argument("--voice-fixer-use-cuda", action='store_true', default=default_arguments['voice-fixer-use-cuda'], help="Hints to voicefixer to use CUDA, if available.")
 	parser.add_argument("--force-cpu-for-conditioning-latents", default=default_arguments['force-cpu-for-conditioning-latents'], action='store_true', help="Forces computing conditional latents to be done on the CPU (if you constantyl OOM on low chunk counts)")
 	parser.add_argument("--device-override", default=default_arguments['device-override'], help="A device string to override pass through Torch")
+	parser.add_argument("--whisper-model", default=default_arguments['whisper-model'], help="Specifies which whisper model to use for transcription.")
 	parser.add_argument("--sample-batch-size", default=default_arguments['sample-batch-size'], type=int, help="Sets how many batches to use during the autoregressive samples pass")
 	parser.add_argument("--concurrency-count", type=int, default=default_arguments['concurrency-count'], help="How many Gradio events to process at once")
 	parser.add_argument("--output-sample-rate", type=int, default=default_arguments['output-sample-rate'], help="Sample rate to resample the output to (from 24KHz)")
@@ -463,7 +465,7 @@ whisper_model = None
 def prepare_dataset( files, outdir ):
 	global whisper_model
 	if whisper_model is None:
-		whisper_model = whisper.load_model("base")
+		whisper_model = whisper.load_model(args.whisper_model)
 
 	os.makedirs(outdir, exist_ok=True)
 
@@ -653,7 +655,7 @@ def get_voice_list(dir=get_voice_dir()):
 	os.makedirs(dir, exist_ok=True)
 	return sorted([d for d in os.listdir(dir) if os.path.isdir(os.path.join(dir, d)) and len(os.listdir(os.path.join(dir, d))) > 0 ]) + ["microphone", "random"]
 
-def export_exec_settings( listen, share, check_for_updates, models_from_local_only, low_vram, embed_output_metadata, latents_lean_and_mean, voice_fixer, voice_fixer_use_cuda, force_cpu_for_conditioning_latents, device_override, sample_batch_size, concurrency_count, output_sample_rate, output_volume ):
+def export_exec_settings( listen, share, check_for_updates, models_from_local_only, low_vram, embed_output_metadata, latents_lean_and_mean, voice_fixer, voice_fixer_use_cuda, force_cpu_for_conditioning_latents, device_override, whisper_model, sample_batch_size, concurrency_count, output_sample_rate, output_volume ):
 	global args
 
 	args.listen = listen
@@ -663,6 +665,7 @@ def export_exec_settings( listen, share, check_for_updates, models_from_local_on
 	args.low_vram = low_vram
 	args.force_cpu_for_conditioning_latents = force_cpu_for_conditioning_latents
 	args.device_override = device_override
+	args.whisper_model = whisper_model
 	args.sample_batch_size = sample_batch_size
 	args.embed_output_metadata = embed_output_metadata
 	args.latents_lean_and_mean = latents_lean_and_mean
@@ -680,6 +683,7 @@ def export_exec_settings( listen, share, check_for_updates, models_from_local_on
 		'models-from-local-only':args.models_from_local_only,
 		'force-cpu-for-conditioning-latents': args.force_cpu_for_conditioning_latents,
 		'device-override': args.device_override,
+		'whisper-model': args.whisper_model,
 		'sample-batch-size': args.sample_batch_size,
 		'embed-output-metadata': args.embed_output_metadata,
 		'latents-lean-and-mean': args.latents_lean_and_mean,

src/webui.py

@@ -370,6 +370,7 @@ def setup_gradio():
 					]
 				)
 		with gr.Tab("Training"):
+			with gr.Tab("Prepare Dataset"):
 				with gr.Row():
 					with gr.Column():
 						dataset_settings = [
@@ -377,6 +378,7 @@ def setup_gradio():
 						]
 						dataset_voices = dataset_settings[0]
 
+					with gr.Column():
 						prepare_dataset_button = gr.Button(value="Prepare")
 
 						def prepare_dataset_proxy( voice ):
@@ -387,7 +389,8 @@ def setup_gradio():
 							inputs=dataset_settings,
 							outputs=None
 						)
-
+			with gr.Tab("Generate Configuration"):
+				with gr.Row():
 					with gr.Column():
 						training_settings = [
 							gr.Slider(label="Batch Size", value=128),
@@ -395,6 +398,7 @@ def setup_gradio():
 							gr.Number(label="Print Frequency", value=50),
 							gr.Number(label="Save Frequency", value=50),
 						]
+						save_yaml_button = gr.Button(value="Save Training Configuration")
 					with gr.Column():
 						training_settings = training_settings + [
 							gr.Textbox(label="Training Name", placeholder="finetune"),
@@ -403,7 +407,7 @@ def setup_gradio():
 							gr.Textbox(label="Validation Name", placeholder="finetune"),
 							gr.Textbox(label="Validation Path", placeholder="./experiments/finetune/val.txt"),
 						]
-						save_yaml_button = gr.Button(value="Save Training Configuration")
+
 						save_yaml_button.click(save_training_settings,
 							inputs=training_settings,
 							outputs=None
@@ -424,6 +428,7 @@ def setup_gradio():
 						gr.Checkbox(label="Use CUDA for Voice Fixer", value=args.voice_fixer_use_cuda),
 						gr.Checkbox(label="Force CPU for Conditioning Latents", value=args.force_cpu_for_conditioning_latents),
 						gr.Textbox(label="Device Override", value=args.device_override),
+						gr.Dropdown(label="Whisper Model", value=args.whisper_model, choices=["tiny", "base", "small", "medium", "large"]),
 					]
 					gr.Button(value="Check for Updates").click(check_for_updates)
 					gr.Button(value="Reload TTS").click(reload_tts)

start.bat

@@ -1,4 +1,5 @@
 call .\venv\Scripts\activate.bat
+set PATH=.\bin\;%PATH%
 python .\src\main.py
 deactivate
 pause