TTS added: Oute & SoVITS; F5 voice change

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text

app/models.py

@@ -45,7 +45,7 @@ AVAILABLE_MODELS = {
     # 'parler-tts/parler-tts-expresso': 'parler-tts/parler-tts-expresso', # 4.29 4.32 4.36.1 4.42.0
 
     # # Microsoft Edge TTS
-    # 'innoai/Edge-TTS-Text-to-Speech': 'innoai/Edge-TTS-Text-to-Speech', # 4.29
+    # 'innoai/Edge-TTS-Text-to-Speech': 'innoai/Edge-TTS-Text-to-Speech', # API disabled
 
     # IMS-Toucan
     # 'Flux9665/MassivelyMultilingualTTS': 'Flux9665/MassivelyMultilingualTTS', # 5.1
@@ -55,10 +55,15 @@ AVAILABLE_MODELS = {
     'hexgrad/kokoro': 'hexgrad/kokoro',
 
     # MaskGCT (by Amphion)
-    # DEMANDS 300 seconds of ZeroGPU
-    # 'amphion/maskgct': 'amphion/maskgct',
-    # default ZeroGPU borrow time
-    'Svngoku/maskgct-audio-lab': 'Svngoku/maskgct-audio-lab',
+    # 'amphion/maskgct': 'amphion/maskgct', # DEMANDS 300 seconds of ZeroGPU!
+    # 'Svngoku/maskgct-audio-lab': 'Svngoku/maskgct-audio-lab', # DEMANDS 300 seconds of ZeroGPU!
+
+    # GPT-SoVITS
+    'lj1995/GPT-SoVITS-v2': 'lj1995/GPT-SoVITS-v2',
+
+    # OuteTTS
+    # 'OuteAI/OuteTTS-0.2-500M-Demo': 'OuteAI/OuteTTS-0.2-500M-Demo',
+    'ameerazam08/OuteTTS-0.2-500M-Demo': 'ameerazam08/OuteTTS-0.2-500M-Demo', # ZeroGPU Space
 
     # HF TTS w issues
     # 'LeeSangHoon/HierSpeech_TTS': 'LeeSangHoon/HierSpeech_TTS', # irresponsive to exclamation marks # 4.29
@@ -230,7 +235,7 @@ HF_SPACES = {
     'Pendrokar/style-tts-2': {
         'name': 'StyleTTS v2',
         'function': '/synthesize',
-        'text_param_index': 0,
+        'text_param_index': 'text',
         'return_audio_index': 0,
         'is_zero_gpu_space': True,
         'series': 'StyleTTS',
@@ -239,12 +244,12 @@ HF_SPACES = {
 
     # StyleTTS v2 kokoro fine tune
     'hexgrad/kokoro': {
-        'name': 'StyleTTS Kokoro',
+        'name': 'StyleTTS Kokoro v19',
         'function': '/generate',
         'text_param_index': 0,
         'return_audio_index': 0,
         'is_zero_gpu_space': True,
-        'series': 'StyleTTS',
+        'series': 'Kokoro',
     },
 
     # MaskGCT (by Amphion)
@@ -266,10 +271,26 @@ HF_SPACES = {
         'series': 'MaskGCT',
         'emoji': '🥵', # 300s minimum ZeroGPU!
     },
+    'lj1995/GPT-SoVITS-v2': {
+        'name': 'GPT-SoVITS',
+        'function': '/get_tts_wav',
+        'text_param_index': 'text',
+        'return_audio_index': 0,
+        'is_zero_gpu_space': True,
+        'series': 'GPT-SoVITS',
+    },
+    'ameerazam08/OuteTTS-0.2-500M-Demo': {
+        'name': 'OuteTTS 500M',
+        'function': '/generate_tts',
+        'text_param_index': 0,
+        'return_audio_index': 0,
+        'is_zero_gpu_space': True,
+        'series': 'OuteTTS',
+    },
 }
 
 # for zero-shot TTS - voice sample used by XTTS (11 seconds)
-DEFAULT_VOICE_SAMPLE_STR = 'https://cdn-uploads.huggingface.co/production/uploads/63d52e0c4e5642795617f668/V6-rMmI-P59DA4leWDIcK.wav'
+DEFAULT_VOICE_SAMPLE_STR = 'voice_samples/xtts_sample.wav'
 DEFAULT_VOICE_SAMPLE = handle_file(DEFAULT_VOICE_SAMPLE_STR)
 DEFAULT_VOICE_TRANSCRIPT = "The Hispaniola was rolling scuppers under in the ocean swell. The booms were tearing at the blocks, the rudder was banging to and fro, and the whole ship creaking, groaning, and jumping like a manufactory."
 
@@ -326,9 +347,12 @@ OVERRIDE_INPUTS = {
         4: 'No', # split by newline
     },
     'mrfakename/MeloTTS': {
-        1: 'EN-Default',	# speaker; DEFAULT_VOICE_SAMPLE=EN-Default
-        2: 1, # speed
-        3: 'EN',	# language
+        # 1: 'EN-Default',	# speaker; DEFAULT_VOICE_SAMPLE=EN-Default
+        # 2: 1, # speed
+        # 3: 'EN',	# language
+        'speaker': 'EN-Default',	# DEFAULT_VOICE_SAMPLE=EN-Default
+        'speed': 1.0,
+        'language': 'EN',
     },
     'mrfakename/MetaVoice-1B-v0.1': {
 		1: 5,	# float (numeric value between 0.0 and 10.0) in 'Speech Stability - improves text following for a challenging speaker' Slider component
@@ -362,13 +386,14 @@ OVERRIDE_INPUTS = {
 		10: "never", #use_memory_cache
     },
 
+    # F5
     'mrfakename/E2-F5-TTS': {
-		0: DEFAULT_VOICE_SAMPLE, # voice sample
-		1: DEFAULT_VOICE_TRANSCRIPT, # transcript of sample (< 15 seconds required)
-		3: False, # cleanup silence
-        4: 0.15, #crossfade
-        5: 32, #nfe_slider
-        6: 1, #speed
+        'ref_audio_input': handle_file('voice_samples/EN_B00004_S00051_W000213.mp3'),
+        'ref_text_input': 'Our model manager is Graham, whom we observed leading a small team of chemical engineers within a multinational European firm we\'ll call Kruger Bern.',
+        'remove_silence': False,
+        'cross_fade_duration_slider': 0.15,
+        'nfe_slider': 32,
+        'speed_slider': 1,
     },
 
     # IMS-Toucan
@@ -383,9 +408,9 @@ OVERRIDE_INPUTS = {
 
     # StyleTTS 2
     'Pendrokar/style-tts-2': {
-		1: "f-us-2", #voice
-        2: 'en-us', # lang
-		3: 8, # lngsteps
+        'voice': "f-us-2",
+        'lang': 'en-us',
+        'lngsteps': 8,
     },
 
     # StyleTTS 2 kokoro
@@ -409,6 +434,29 @@ OVERRIDE_INPUTS = {
 		2: -1, #target_len
 		3: 25, #n_timesteps
     },
+    'lj1995/GPT-SoVITS-v2': {
+        'ref_wav_path': handle_file('voice_samples/EN_B00004_S00051_W000213.wav'),
+        'prompt_text': "Our model manager is Graham, whom we observed leading a small team of chemical engineers within a multinational European firm we'll call",
+        'prompt_language': "English",
+        # text: "Please surprise me and speak in whatever voice you enjoy.",
+        'text_language': "English",
+        'how_to_cut': "No slice",
+        'top_k': 15,
+        'top_p': 1,
+        'temperature': 1,
+        'ref_free': False,
+        'speed': 1,
+        'if_freeze': False,
+        'inp_refs': None,
+    },
+    'ameerazam08/OuteTTS-0.2-500M-Demo': {
+        1: 0.1, # temperature
+        2: 1.1, # repetition_penalty
+        3: "en", # language
+        4: "female_1", # speaker_selection
+        5: None, # reference_audio
+        6: None, # reference_text
+    },
 }
 
 

app/synth.py

@@ -135,7 +135,19 @@ def synthandreturn(text, autoplay, request: gr.Request):
                         space_inputs[HF_SPACES[model]['text_param_index']] = text
 
                         print(f"{model}: Sending request to HF Space")
-                        results = mdl_space.predict(*space_inputs, api_name=api_name, fn_index=fn_index)
+                        # results = mdl_space.predict(*space_inputs, api_name=api_name, fn_index=fn_index)
+                        if(type(space_inputs) == dict):
+                            results = mdl_space.predict(
+                                **space_inputs,
+                                api_name=api_name,
+                                fn_index=fn_index
+                            )
+                        else:
+                            results = mdl_space.predict(
+                                *space_inputs,
+                                api_name=api_name,
+                                fn_index=fn_index
+                            )
 
                         # return path to audio
                         result = results
@@ -189,31 +201,64 @@ def synthandreturn(text, autoplay, request: gr.Request):
         result_storage[model] = result
 
     def _get_param_examples(parameters):
-        example_inputs = []
+        # named or unnamed parameters
+        try:
+            param_name = parameters[0]['parameter_name']
+            # success => named params, use dict
+            example_inputs = {}
+        except:
+            # unnamed params, use list
+            example_inputs = []
+            pass
+
         for param_info in parameters:
+
+
+            param_name = ''
+            param_default_value = param_info['example_input']
+            try:
+                # named params
+                param_name = param_info['parameter_name']
+                param_default_value = param_info['parameter_default']
+            except:
+                # unnamed params
+                pass
+
+            param_value = None
             if (
                 param_info['component'] == 'Radio'
                 or param_info['component'] == 'Dropdown'
                 or param_info['component'] == 'Audio'
                 or param_info['python_type']['type'] == 'str'
             ):
-                example_inputs.append(str(param_info['example_input']))
-                continue
-            if param_info['python_type']['type'] == 'int':
-                example_inputs.append(int(param_info['example_input']))
-                continue
-            if param_info['python_type']['type'] == 'float':
-                example_inputs.append(float(param_info['example_input']))
-                continue
-            if param_info['python_type']['type'] == 'bool':
-                example_inputs.append(bool(param_info['example_input']))
-                continue
+                param_value = str(param_default_value)
+            elif param_info['python_type']['type'] == 'int':
+                param_value = int(param_default_value)
+            elif param_info['python_type']['type'] == 'float':
+                param_value = float(param_default_value)
+            elif param_info['python_type']['type'] == 'bool':
+                param_value = bool(param_default_value)
+
+            if (param_name != ''):
+                # named param
+                example_inputs[param_info['parameter_name']] = param_value
+            else:
+                # just append unnamed param and hope
+                example_inputs.append(param_value)
 
         return example_inputs
 
     def _override_params(inputs, modelname):
         try:
             for key,value in OVERRIDE_INPUTS[modelname].items():
+                # if override keys are integers, make the dict into a list
+                if (
+                    (type(inputs) is dict)
+                    and (type(key) is int)
+                ):
+                    print(f"{modelname}: Converting unnamed override params to List")
+                    inputs = list(inputs.values())
+
                 inputs[key] = value
             print(f"{modelname}: Default inputs overridden by Arena")
         except:

test_overrides.py

@@ -0,0 +1,84 @@
+from app.models import *
+
+# # has named endpoint
+# if '/' == HF_SPACES[model]['function'][0]:
+#     # audio sync function name
+#     api_name = HF_SPACES[model]['function']
+
+#     end_parameters = _get_param_examples(
+#         endpoints['named_endpoints'][api_name]['parameters']
+#     )
+# # has unnamed endpoint
+# else:
+#     # endpoint index is the first character
+#     fn_index = int(HF_SPACES[model]['function'])
+
+#     end_parameters = _get_param_examples(
+#         endpoints['unnamed_endpoints'][str(fn_index)]['parameters']
+#     )
+
+def _get_param_examples(parameters):
+    # named or unnamed parameters
+    try:
+        param_name = parameters[0]['parameter_name']
+        # success => named params, use dict
+        example_inputs = {}
+    except:
+        # unnamed params, use list
+        example_inputs = []
+        pass
+
+    for param_info in parameters:
+
+
+        param_name = ''
+        param_default_value = param_info['example_input']
+        try:
+            # named params
+            param_name = param_info['parameter_name']
+            param_default_value = param_info['parameter_default']
+        except:
+            # unnamed params
+            pass
+
+        param_value = None
+        if (
+            param_info['component'] == 'Radio'
+            or param_info['component'] == 'Dropdown'
+            or param_info['component'] == 'Audio'
+            or param_info['python_type']['type'] == 'str'
+        ):
+            param_value = str(param_default_value)
+        elif param_info['python_type']['type'] == 'int':
+            param_value = int(param_default_value)
+        elif param_info['python_type']['type'] == 'float':
+            param_value = float(param_default_value)
+        elif param_info['python_type']['type'] == 'bool':
+            param_value = bool(param_default_value)
+
+        if (param_name != ''):
+            # named param
+            example_inputs[param_info['parameter_name']] = param_value
+        else:
+            # just append unnamed param and hope
+            example_inputs.append(param_value)
+
+    return example_inputs
+
+def _override_params(inputs, modelname):
+    try:
+        for key,value in OVERRIDE_INPUTS[modelname].items():
+            # if override keys are integers, make the dict into a list
+            if (
+                (type(inputs) is dict)
+                and (type(key) is int)
+            ):
+                print("Converting unnamed override params to List")
+                inputs = list(inputs.values())
+
+            inputs[key] = value
+        print(f"{modelname}: Default inputs overridden by Arena")
+    except:
+        pass
+
+    return inputs

test_tts_cosyvoice.py

@@ -0,0 +1,19 @@
+import os
+from gradio_client import Client, handle_file
+
+client = Client("FunAudioLLM/CosyVoice2-0.5B", hf_token=os.getenv('HF_TOKEN'))
+endpoints = client.view_api(all_endpoints=True, print_info=False, return_format='dict')
+# print(endpoints)
+
+result = client.predict(
+		tts_text="CosyVoice is undergoing a comprehensive upgrade, providing more accurate, stable, faster, and better voice generation capabilities.",
+		mode_checkbox_group="3s Voice Clone",
+		prompt_text='The Hispaniola was rolling scuppers under in the ocean swell. The booms were tearing at the blocks, the rudder was banging to and fro, and the whole ship creaking, groaning, and jumping like a manufactory.',
+		prompt_wav_upload=handle_file("https://cdn-uploads.huggingface.co/production/uploads/63d52e0c4e5642795617f668/V6-rMmI-P59DA4leWDIcK.wav"),
+		prompt_wav_record=None,
+		instruct_text=None,
+		seed=0,
+		stream="false",
+		api_name="/generate_audio"
+)
+print(result)

test_tts_e2_f5_f5.py

@@ -4,7 +4,7 @@ from gradio_client import Client, handle_file
 client = Client("mrfakename/E2-F5-TTS", hf_token=os.getenv('HF_TOKEN'))
 endpoints = client.view_api(all_endpoints=True, print_info=False, return_format='dict')
 result = client.predict(
-		ref_audio_input=handle_file('https://cdn-uploads.huggingface.co/production/uploads/63d52e0c4e5642795617f668/V6-rMmI-P59DA4leWDIcK.wav'),
+		ref_audio_input=handle_file('voice_samples/EN_B00004_S00051_W000213.mp3'),
 		ref_text_input="The Hispaniola was rolling scuppers under in the ocean swell. The booms were tearing at the blocks, the rudder was banging to and fro, and the whole ship creaking, groaning, and jumping like a manufactory.",
 		gen_text_input="Please surprise me and speak in whatever voice you enjoy.",
 		remove_silence=False,

test_tts_melo.py

@@ -1,13 +1,42 @@
 import os
+from test_overrides import _get_param_examples, _override_params
 from gradio_client import Client
 
-client = Client("mrfakename/MeloTTS", hf_token=os.getenv('HF_TOKEN'))
+model = "mrfakename/MeloTTS"
+client = Client(model, hf_token=os.getenv('HF_TOKEN'))
 endpoints = client.view_api(all_endpoints=True, print_info=False, return_format='dict')
 # print(endpoints)
-result = client.predict(
-		"Please surprise me and speak in whatever voice you enjoy.",	# str in 'Text to speak' Textbox component
-		"EN-US",	# Literal['EN-US', 'EN-BR', 'EN_INDIA', 'EN-AU', 'EN-Default'] in 'Speaker' Dropdown component
-		1.0,	# float (numeric value between 0.1 and 10.0)
-		"EN",	# Literal['EN', 'ES', 'FR', 'ZH', 'JP', 'KR'] in 'Language' Radio component
-		api_name="/synthesize"
-)
+
+api_name = '/synthesize'
+fn_index = None
+end_parameters = None
+text = 'This is what my voice sounds like.'
+
+end_parameters = _get_param_examples(
+	endpoints['named_endpoints'][api_name]['parameters']
+)
+print(end_parameters)
+
+
+# override some or all default parameters
+space_inputs = _override_params(end_parameters, model)
+
+# space_inputs[0] = text
+space_inputs['text'] = text
+print(space_inputs)
+
+if(type(space_inputs) == dict):
+	space_inputs['text'] = text
+	result = client.predict(
+		**space_inputs,
+		api_name=api_name,
+		fn_index=fn_index
+	)
+else:
+	space_inputs[0] = text
+	result = client.predict(
+		*space_inputs,
+		api_name=api_name,
+		fn_index=fn_index
+	)
+print(result)

test_tts_oute.py

@@ -0,0 +1,22 @@
+import os
+from gradio_client import Client, handle_file
+
+# client = Client("OuteAI/OuteTTS-0.2-500M-Demo", hf_token=os.getenv('HF_TOKEN'))
+client = Client("ameerazam08/OuteTTS-0.2-500M-Demo", hf_token=os.getenv('HF_TOKEN'))
+
+endpoints = client.view_api(all_endpoints=True, print_info=False, return_format='dict')
+# print(endpoints)
+
+result = client.predict(
+		text="Please surprise me and speak in whatever voice you enjoy.",
+		temperature=0.1,
+		repetition_penalty=1.1,
+		language="en",
+		speaker_selection="female_1",
+		reference_audio=None,
+		reference_text=None,
+		# reference_audio=handle_file('EN_B00004_S00051_W000213.wav'),
+		# reference_text="Our model manager is Graham, whom we observed leading a small team of chemical engineers within a multinational European firm we'll call",
+		api_name="/generate_tts"
+)
+print(result)

test_tts_sovits.py

@@ -0,0 +1,46 @@
+import os
+from test_overrides import _get_param_examples, _override_params
+from gradio_client import Client, handle_file
+
+model = "Pendrokar/GPT-SoVITS-v2"
+# lj1995/GPT-SoVITS-v2
+client = Client(model, hf_token=os.getenv('HF_TOKEN'))
+endpoints = client.view_api(all_endpoints=True, print_info=False, return_format='dict')
+# print(endpoints)
+
+
+api_name = None
+fn_index = None
+end_parameters = None
+text = 'This is what my voice sounds like.'
+
+# has named endpoint
+# audio sync function name
+api_name = '/get_tts_wav'
+
+end_parameters = _get_param_examples(
+	endpoints['named_endpoints'][api_name]['parameters']
+)
+print(end_parameters)
+
+# override some or all default parameters
+space_inputs = _override_params(end_parameters, model)
+
+print(space_inputs)
+
+if(type(space_inputs) == dict):
+	space_inputs['text'] = text
+	result = client.predict(
+		**space_inputs,
+		api_name=api_name,
+		fn_index=fn_index
+	)
+else:
+	space_inputs[0] = text
+	result = client.predict(
+		*space_inputs,
+		api_name=api_name,
+		fn_index=fn_index
+	)
+
+print(result)

test_tts_styletts.py

@@ -1,12 +1,50 @@
 import os
+from test_overrides import _get_param_examples, _override_params
 from gradio_client import Client, file
 
-client = Client("Pendrokar/style-tts-2", hf_token=os.getenv('HF_TOKEN'))
+model = "Pendrokar/style-tts-2"
+client = Client(model, hf_token=os.getenv('HF_TOKEN'))
 endpoints = client.view_api(all_endpoints=True, print_info=False, return_format='dict')
 # print(endpoints)
-result = client.predict(
-		text="Hello!!",
-		voice="f-us-1", # voice
-		lngsteps=8, # lngsteps
-		api_name="/synthesize" # api_name
-)
+
+api_name = '/synthesize'
+fn_index = None
+end_parameters = None
+text = 'This is what my voice sounds like.'
+
+end_parameters = _get_param_examples(
+	endpoints['named_endpoints'][api_name]['parameters']
+)
+print(end_parameters)
+
+
+space_inputs = end_parameters
+# override some or all default parameters
+space_inputs = _override_params(end_parameters, model)
+
+if(type(space_inputs) == dict):
+	space_inputs['text'] = text
+	result = client.predict(
+		**space_inputs,
+		api_name=api_name,
+		fn_index=fn_index
+	)
+else:
+	space_inputs[0] = text
+	result = client.predict(
+		*space_inputs,
+		api_name=api_name,
+		fn_index=fn_index
+	)
+	# space_inputs = {str(i): value for i, value in enumerate(space_inputs)}
+
+print(space_inputs)
+# print(*space_inputs)
+# print(**space_inputs)
+
+# result = client.predict(
+# 	**space_inputs,
+# 	api_name=api_name,
+#     fn_index=fn_index
+# )
+print(result)

test_tts_xva.py

@@ -1,23 +1,30 @@
 import os
+from test_overrides import _get_param_examples, _override_params
 from gradio_client import Client, file
 
+model = "Pendrokar/xVASynth-TTS"
 client = Client("Pendrokar/xVASynth-TTS", hf_token=os.getenv('HF_TOKEN'))
 endpoints = client.view_api(all_endpoints=True, print_info=False, return_format='dict')
-# print(endpoints)
-result = client.predict(
-		"Well, hello there!!",	# str  in 'Input Text' Textbox component
-		"x_ex04",	# Literal['x_ex04', 'x_ex01', 'cnc_cabal', 'ccby_nvidia_hifi_92_F', 'ccby_nvidia_hifi_6671_M', 'more']  in 'Voice' Radio component
-		"en",	# Literal['en', 'de', 'es', 'hi', 'zh', 'more']  in 'Language' Radio component
-		1.0,	# float (numeric value between 0.5 and 2.0) in 'Duration' Slider component
 
-		0,	# UNUSED; float (numeric value between 0 and 1.0) in 'Pitch' Slider component
-		0.1,	# UNUSED; float (numeric value between 0.1 and 1.0) in 'Energy' Slider component
+api_name = '/predict'
+fn_index = None
+end_parameters = None
+text = 'This is what my voice sounds like.'
+
+end_parameters = _get_param_examples(
+	endpoints['named_endpoints'][api_name]['parameters']
+)
+print(end_parameters)
+
 
-		0,	# Overriden by DeepMoji; float (numeric value between 0 and 1.0) in '😠 Anger' Slider component
-		0,	# Overriden by DeepMoji; float (numeric value between 0 and 1.0) in '😃 Happiness' Slider component
-		0,	# Overriden by DeepMoji; float (numeric value between 0 and 1.0) in '😭 Sadness' Slider component
-		0,	# Overriden by DeepMoji; float (numeric value between 0 and 1.0) in '😮 Surprise' Slider component
-		True,	# bool  in 'Use DeepMoji' Checkbox component
+# override some or all default parameters
+space_inputs = _override_params(end_parameters, model)
 
-		api_name="/predict"
-)
+space_inputs[0] = text
+
+print(space_inputs)
+result = client.predict(
+	*space_inputs,
+	api_name=api_name
+)
+print(result)

voice_samples/EN_B00004_S00051_W000213.json

@@ -0,0 +1 @@
+{"id": "EN_B00004_S00051_W000213", "wav": "EN_B00004/EN_B00004_S00051/mp3/EN_B00004_S00051_W000213.mp3", "text": " Our model manager is Graham, whom we observed leading a small team of chemical engineers within a multinational European firm we'll call Kruger Bern.", "duration": 10.1535, "speaker": "EN_B00004_S00051", "language": "en", "dnsmos": 3.3549}