tweaks, including exporting on save/quit

vall_e/config.py

@@ -372,6 +372,10 @@ class Trainer:
 
 	save_on_oom: bool = True
 	save_on_quit: bool = True
+	
+	export_on_save: bool = False
+	export_on_quit: bool = False
+	
 	save_frequency: int = 100
 
 	keep_last_checkpoints: int = 0

vall_e/data.py

@@ -10,7 +10,7 @@ import random
 import torch
 
 from .config import cfg
-from .emb.qnt import trim_random, repeat_extend_audio, merge_audio, decode_to_file
+from .emb.qnt import trim, trim_random, repeat_extend_audio, merge_audio, decode_to_file
 
 from collections import defaultdict
 from functools import cache, cached_property
@@ -253,7 +253,7 @@ class Dataset(_Dataset):
 				qnt = _load_quants(path)
 
 			if cfg.dataset.prompt_duration > 0 and trim_length < qnt.shape[0]:
-				qnt = trim_random( qnt, trim_length )
+				qnt = trim( qnt, trim_length )
 
 			prom_list.append(qnt)
 			prom_length += qnt.shape[0]
@@ -264,7 +264,7 @@ class Dataset(_Dataset):
 		prom = torch.cat(prom_list)
 
 		if cfg.dataset.prompt_duration > 0 and trim_length < prom.shape[0]:
-			prom = trim_random( prom, trim_length )
+			prom = trim( prom, trim_length )
 
 		return prom
 

vall_e/emb/qnt.py

@@ -180,7 +180,19 @@ def encode_from_file(path, device="cuda"):
 
 # Helper Functions
 
+# trims from the start, up to `target`
+def trim( qnt, target ):
+	length = max( qnt.shape[0], qnt.shape[1] )
+	start = 0
+	end = start + target
+	if end >= length:
+		start = length - target
+		end = length                
+
+	return qnt[start:end] if qnt.shape[0] > qnt.shape[1] else qnt[:, start:end]
+
 # trims a random piece of audio, up to `target`
+# to-do: try and align to EnCodec window
 def trim_random( qnt, target ):
 	length = max( qnt.shape[0], qnt.shape[1] )
 	start = int(length * random.random())

vall_e/inference.py

@@ -7,7 +7,7 @@ from einops import rearrange
 from pathlib import Path
 
 from .emb import g2p, qnt
-from .emb.qnt import trim_random
+from .emb.qnt import trim, trim_random
 from .utils import to_device
 
 from .config import cfg
@@ -25,12 +25,14 @@ class TTS():
 
 		if config:
 			cfg.load_yaml( config )
+			cfg.dataset.use_hdf5 = False # could use cfg.load_hdf5(), but why would it ever need to be loaded for inferencing
 
 		try:
 			cfg.format()
 		except Exception as e:
 			pass
 		
+		self.symmap = None
 		if ar_ckpt and nar_ckpt:
 			self.ar_ckpt = ar_ckpt
 			self.nar_ckpt = nar_ckpt
@@ -40,6 +42,8 @@ class TTS():
 				if name.startswith("ar"):
 					self.ar = model
 					state = torch.load(self.ar_ckpt)
+					if "symmap" in state:
+						self.symmap = state['symmap']
 					if "module" in state:
 						state = state['module']
 					self.ar.load_state_dict(state)
@@ -47,6 +51,8 @@ class TTS():
 				elif name.startswith("nar"):
 					self.nar = model
 					state = torch.load(self.nar_ckpt)
+					if "symmap" in state:
+						self.symmap = state['symmap']
 					if "module" in state:
 						state = state['module']
 					self.nar.load_state_dict(state)
@@ -54,7 +60,9 @@ class TTS():
 		else:
 			self.load_models()
 
-		self.symmap = get_phone_symmap()
+		if self.symmap is None:
+			self.symmap = get_phone_symmap()
+
 		self.ar.eval()
 		self.nar.eval()
 
@@ -78,7 +86,7 @@ class TTS():
 		phones = [ " " if not p else p for p in content ]
 		return torch.tensor([ 1 ] + [*map(self.symmap.get, phones)] + [ 2 ])
 
-	def encode_audio( self, paths, trim=True ):
+	def encode_audio( self, paths, should_trim=True ):
 		# already a tensor, return it
 		if isinstance( paths, Tensor ):
 			return paths
@@ -90,7 +98,7 @@ class TTS():
 		# merge inputs
 		res = torch.cat([qnt.encode_from_file( path )[0].t().to(torch.int16) for path in paths])
 
-		if trim:
+		if should_trim:
 			res = trim_random( res, int( 75 * cfg.dataset.prompt_duration ) )
 		
 		return res

vall_e/utils/trainer.py

@@ -324,14 +324,24 @@ def train(
 			save_ckpt_every = cfg.trainer.save_frequency or cfg.evaluation.frequency
 
 			saving_commands = ["save"]
+			export_commands = ["export"]
 
 			if cfg.trainer.save_on_quit:
 				saving_commands.append("quit")
 
+			if cfg.trainer.export_on_quit:
+				export_commands.append("quit")
+
+			if cfg.trainer.export_on_save:
+				export_commands.append("save")
+
 			if engines.global_step != last_save_step:
 				if engines.global_step % save_ckpt_every == 0 or command in saving_commands:
 					engines.save_checkpoint()
 					last_save_step = engines.global_step
+					
+					if command in export_commands and is_global_leader():
+						engines.export(userdata={"symmap": get_phone_symmap()})
 
 			if engines.global_step != last_eval_step:
 				if engines.global_step % cfg.evaluation.frequency == 0 or command in ["eval"]:
@@ -343,4 +353,5 @@ def train(
 					last_eval_step = engines.global_step
 
 			if command in ["quit"]:
+				if cfg.export_on_quit:
 				return