audio-while-streaming

README.md

@@ -4,11 +4,11 @@ emoji: 🐸
 colorFrom: green
 colorTo: red
 sdk: gradio
-sdk_version: 3.48.0
+sdk_version: 3.44.3
 app_file: app.py
 pinned: false
 models:
-- coqui/XTTS-v2
+- coqui/XTTS-v1
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -3,26 +3,16 @@ import io, os, stat
 import subprocess
 import random
 from zipfile import ZipFile
-import uuid
+import uuid 
 import time
 import torch
 import torchaudio
-
-
-#download for mecab
-os.system('python -m unidic download')
-
 # By using XTTS you agree to CPML license https://coqui.ai/cpml
 os.environ["COQUI_TOS_AGREED"] = "1"
 
 # langid is used to detect language for longer text
 # Most users expect text to be their own language, there is checkbox to disable it
-import langid
-import base64
-import csv
-from io import StringIO
-import datetime
-import re
+import langid 
 
 import gradio as gr
 from scipy.io.wavfile import write
@@ -45,215 +35,153 @@ repo_id = "coqui/xtts"
 print("Export newer ffmpeg binary for denoise filter")
 ZipFile("ffmpeg.zip").extractall()
 print("Make ffmpeg binary executable")
-st = os.stat("ffmpeg")
-os.chmod("ffmpeg", st.st_mode | stat.S_IEXEC)
+st = os.stat('ffmpeg')
+os.chmod('ffmpeg', st.st_mode | stat.S_IEXEC)
 
-# This will trigger downloading model
-print("Downloading if not downloaded Coqui XTTS V2")
-from TTS.utils.manage import ModelManager
-
-model_name = "tts_models/multilingual/multi-dataset/xtts_v2"
-ModelManager().download_model(model_name)
-model_path = os.path.join(get_user_data_dir("tts"), model_name.replace("/", "--"))
-print("XTTS downloaded")
+# Load TTS
+tts = TTS("tts_models/multilingual/multi-dataset/xtts_v1")
 
+model_path = os.path.join(get_user_data_dir("tts"), "tts_models--multilingual--multi-dataset--xtts_v1")
 config = XttsConfig()
 config.load_json(os.path.join(model_path, "config.json"))
-
 model = Xtts.init_from_config(config)
 model.load_checkpoint(
     config,
     checkpoint_path=os.path.join(model_path, "model.pth"),
     vocab_path=os.path.join(model_path, "vocab.json"),
     eval=True,
-    use_deepspeed=True,
+    use_deepspeed=True
 )
 model.cuda()
 
 # This is for debugging purposes only
-DEVICE_ASSERT_DETECTED = 0
-DEVICE_ASSERT_PROMPT = None
-DEVICE_ASSERT_LANG = None
-
-supported_languages = config.languages
+DEVICE_ASSERT_DETECTED=0
+DEVICE_ASSERT_PROMPT=None
+DEVICE_ASSERT_LANG=None
 
-def predict(
-    prompt,
-    language,
-    audio_file_pth,
-    mic_file_path,
-    use_mic,
-    voice_cleanup,
-    no_lang_auto_detect,
-    agree,
-):
+def predict(prompt, language, audio_file_pth, mic_file_path, use_mic, voice_cleanup, no_lang_auto_detect, agree,):
     if agree == True:
+        supported_languages=["en","es","fr","de","it","pt","pl","tr","ru","nl","cs","ar","zh-cn"]
+        
         if language not in supported_languages:
-            gr.Warning(
-                f"Language you put {language} in is not in is not in our Supported Languages, please choose from dropdown"
-            )
-
+            gr.Warning(f"Language you put {language} in is not in is not in our Supported Languages, please choose from dropdown")
+                
             return (
-                None,
-                None,
-                None,
-                None,
-            )
+                    None,
+                    None,
+                    None,
+                    None,
+                ) 
 
-        language_predicted = langid.classify(prompt)[
-            0
-        ].strip()  # strip need as there is space at end!
+        language_predicted=langid.classify(prompt)[0].strip() # strip need as there is space at end!
 
         # tts expects chinese as zh-cn
-        if language_predicted == "zh":
-            # we use zh-cn
+        if language_predicted == "zh": 
+            #we use zh-cn 
             language_predicted = "zh-cn"
-
         print(f"Detected language:{language_predicted}, Chosen language:{language}")
 
         # After text character length 15 trigger language detection
-        if len(prompt) > 15:
+        if len(prompt)>15:
             # allow any language for short text as some may be common
             # If user unchecks language autodetection it will not trigger
             # You may remove this completely for own use
             if language_predicted != language and not no_lang_auto_detect:
-                # Please duplicate and remove this check if you really want this
-                # Or auto-detector fails to identify language (which it can on pretty short text or mixed text)
-                gr.Warning(
-                    f"It looks like your text isn’t the language you chose , if you’re sure the text is the same language you chose, please check disable language auto-detection checkbox"
-                )
-
+                #Please duplicate and remove this check if you really want this
+                #Or auto-detector fails to identify language (which it can on pretty short text or mixed text)
+                gr.Warning(f"It looks like your text isn’t the language you chose , if you’re sure the text is the same language you chose, please check disable language auto-detection checkbox" )
+            
                 return (
-                    None,
-                    None,
-                    None,
-                    None,
-                )
+                        None,
+                        None,
+                        None,
+                        None,
+                    ) 
 
+        
         if use_mic == True:
             if mic_file_path is not None:
-                speaker_wav = mic_file_path
+               speaker_wav=mic_file_path
             else:
-                gr.Warning(
-                    "Please record your voice with Microphone, or uncheck Use Microphone to use reference audios"
-                )
+                gr.Warning("Please record your voice with Microphone, or uncheck Use Microphone to use reference audios")
                 return (
                     None,
                     None,
                     None,
                     None,
-                )
-
+                ) 
+                
         else:
-            speaker_wav = audio_file_pth
+            speaker_wav=audio_file_pth
 
+        
         # Filtering for microphone input, as it has BG noise, maybe silence in beginning and end
         # This is fast filtering not perfect
 
         # Apply all on demand
-        lowpassfilter = denoise = trim = loudness = True
-
+        lowpassfilter=denoise=trim=loudness=True
+        
         if lowpassfilter:
-            lowpass_highpass = "lowpass=8000,highpass=75,"
+            lowpass_highpass="lowpass=8000,highpass=75," 
         else:
-            lowpass_highpass = ""
+            lowpass_highpass=""
 
         if trim:
             # better to remove silence in beginning and end for microphone
-            trim_silence = "areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,"
+            trim_silence="areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,"
         else:
-            trim_silence = ""
-
-        if voice_cleanup:
+            trim_silence=""
+            
+        if (voice_cleanup):
             try:
-                out_filename = (
-                    speaker_wav + str(uuid.uuid4()) + ".wav"
-                )  # ffmpeg to know output format
-
-                # we will use newer ffmpeg as that has afftn denoise filter
-                shell_command = f"./ffmpeg -y -i {speaker_wav} -af {lowpass_highpass}{trim_silence} {out_filename}".split(
-                    " "
-                )
-
-                command_result = subprocess.run(
-                    [item for item in shell_command],
-                    capture_output=False,
-                    text=True,
-                    check=True,
-                )
-                speaker_wav = out_filename
+                out_filename = speaker_wav + str(uuid.uuid4()) + ".wav"  #ffmpeg to know output format
+    
+                #we will use newer ffmpeg as that has afftn denoise filter
+                shell_command = f"./ffmpeg -y -i {speaker_wav} -af {lowpass_highpass}{trim_silence} {out_filename}".split(" ")
+    
+                command_result = subprocess.run([item for item in shell_command], capture_output=False,text=True, check=True)
+                speaker_wav=out_filename
                 print("Filtered microphone input")
             except subprocess.CalledProcessError:
                 # There was an error - command exited with non-zero code
                 print("Error: failed filtering, use original microphone input")
         else:
-            speaker_wav = speaker_wav
+            speaker_wav=speaker_wav
 
-        if len(prompt) < 2:
+        if len(prompt)<2:
             gr.Warning("Please give a longer prompt text")
             return (
-                None,
-                None,
-                None,
-                None,
-            )
-        if len(prompt) > 200:
-            gr.Warning(
-                "Text length limited to 200 characters for this demo, please try shorter text. You can clone this space and edit code for your own usage"
-            )
+                    None,
+                    None,
+                    None,
+                    None,
+                )
+        if len(prompt)>200:
+            gr.Warning("Text length limited to 200 characters for this demo, please try shorter text. You can clone this space and edit code for your own usage")
             return (
-                None,
-                None,
-                None,
-                None,
-            )
+                    None,
+                    None,
+                    None,
+                    None,
+                )  
         global DEVICE_ASSERT_DETECTED
         if DEVICE_ASSERT_DETECTED:
             global DEVICE_ASSERT_PROMPT
             global DEVICE_ASSERT_LANG
-            # It will likely never come here as we restart space on first unrecoverable error now
-            print(
-                f"Unrecoverable exception caused by language:{DEVICE_ASSERT_LANG} prompt:{DEVICE_ASSERT_PROMPT}"
-            )
-
-            # HF Space specific.. This error is unrecoverable need to restart space
-            space = api.get_space_runtime(repo_id=repo_id)
-            if space.stage!="BUILDING":
-                api.restart_space(repo_id=repo_id)
-            else:
-                print("TRIED TO RESTART but space is building")
-
-        try:
-            metrics_text = ""
-            t_latent = time.time()
-
+            #It will likely never come here as we restart space on first unrecoverable error now
+            print(f"Unrecoverable exception caused by language:{DEVICE_ASSERT_LANG} prompt:{DEVICE_ASSERT_PROMPT}")
+            
+        try:   
+            metrics_text=""
+            t_latent=time.time()
+            
             # note diffusion_conditioning not used on hifigan (default mode), it will be empty but need to pass it to model.inference
-            try:
-                (
-                    gpt_cond_latent,
-                    speaker_embedding,
-                ) = model.get_conditioning_latents(audio_path=speaker_wav, gpt_cond_len=30, gpt_cond_chunk_len=4, max_ref_length=60)
-            except Exception as e:
-                print("Speaker encoding error", str(e))
-                gr.Warning(
-                    "It appears something wrong with reference, did you unmute your microphone?"
-                )
-                return (
-                    None,
-                    None,
-                    None,
-                    None,
-                )
-
+            gpt_cond_latent, diffusion_conditioning, speaker_embedding = model.get_conditioning_latents(audio_path=speaker_wav)
             latent_calculation_time = time.time() - t_latent
-            # metrics_text=f"Embedding calculation time: {latent_calculation_time:.2f} seconds\n"
-
-            # temporary comma fix
-            prompt= re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)",r"\1 \2\2",prompt)
-
-            wav_chunks = []
-            ## Direct mode
+            #metrics_text=f"Embedding calculation time: {latent_calculation_time:.2f} seconds\n"
             
+            wav_chunks = []
+    
             print("I: Generating new audio...")
             t0 = time.time()
             out = model.inference(
@@ -261,8 +189,7 @@ def predict(
                 language,
                 gpt_cond_latent,
                 speaker_embedding,
-                repetition_penalty=5.0,
-                temperature=0.75,
+                diffusion_conditioning
             )
             inference_time = time.time() - t0
             print(f"I: Time to generate audio: {round(inference_time*1000)} milliseconds")
@@ -271,125 +198,24 @@ def predict(
             print(f"Real-time factor (RTF): {real_time_factor}")
             metrics_text+=f"Real-time factor (RTF): {real_time_factor:.2f}\n"
             torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
-
-
-            """
-            print("I: Generating new audio in streaming mode...")
-            t0 = time.time()
-            chunks = model.inference_stream(
-                prompt,
-                language,
-                gpt_cond_latent,
-                speaker_embedding,
-                repetition_penalty=7.0,
-                temperature=0.85,
-            )
-
-            first_chunk = True
-            for i, chunk in enumerate(chunks):
-                if first_chunk:
-                    first_chunk_time = time.time() - t0
-                    metrics_text += f"Latency to first audio chunk: {round(first_chunk_time*1000)} milliseconds\n"
-                    first_chunk = False
-                wav_chunks.append(chunk)
-                print(f"Received chunk {i} of audio length {chunk.shape[-1]}")
-            inference_time = time.time() - t0
-            print(
-                f"I: Time to generate audio: {round(inference_time*1000)} milliseconds"
-            )
-            #metrics_text += (
-            #    f"Time to generate audio: {round(inference_time*1000)} milliseconds\n"
-            #)
-
-            wav = torch.cat(wav_chunks, dim=0)
-            print(wav.shape)
-            real_time_factor = (time.time() - t0) / wav.shape[0] * 24000
-            print(f"Real-time factor (RTF): {real_time_factor}")
-            metrics_text += f"Real-time factor (RTF): {real_time_factor:.2f}\n"
-
-            torchaudio.save("output.wav", wav.squeeze().unsqueeze(0).cpu(), 24000)
-            """
-
-        except RuntimeError as e:
+            
+        except RuntimeError as e :
             if "device-side assert" in str(e):
                 # cannot do anything on cuda device side error, need tor estart
-                print(
-                    f"Exit due to: Unrecoverable exception caused by language:{language} prompt:{prompt}",
-                    flush=True,
-                )
+                print(f"Exit due to: Unrecoverable exception caused by language:{language} prompt:{prompt}", flush=True)
                 gr.Warning("Unhandled Exception encounter, please retry in a minute")
                 print("Cuda device-assert Runtime encountered need restart")
                 if not DEVICE_ASSERT_DETECTED:
-                    DEVICE_ASSERT_DETECTED = 1
-                    DEVICE_ASSERT_PROMPT = prompt
-                    DEVICE_ASSERT_LANG = language
-
-                # just before restarting save what caused the issue so we can handle it in future
-                # Uploading Error data only happens for unrecovarable error
-                error_time = datetime.datetime.now().strftime("%d-%m-%Y-%H:%M:%S")
-                error_data = [
-                    error_time,
-                    prompt,
-                    language,
-                    audio_file_pth,
-                    mic_file_path,
-                    use_mic,
-                    voice_cleanup,
-                    no_lang_auto_detect,
-                    agree,
-                ]
-                error_data = [str(e) if type(e) != str else e for e in error_data]
-                print(error_data)
-                print(speaker_wav)
-                write_io = StringIO()
-                csv.writer(write_io).writerows([error_data])
-                csv_upload = write_io.getvalue().encode()
-
-                filename = error_time + "_" + str(uuid.uuid4()) + ".csv"
-                print("Writing error csv")
-                error_api = HfApi()
-                error_api.upload_file(
-                    path_or_fileobj=csv_upload,
-                    path_in_repo=filename,
-                    repo_id="coqui/xtts-flagged-dataset",
-                    repo_type="dataset",
-                )
+                    DEVICE_ASSERT_DETECTED=1
+                    DEVICE_ASSERT_PROMPT=prompt
+                    DEVICE_ASSERT_LANG=language
 
-                # speaker_wav
-                print("Writing error reference audio")
-                speaker_filename = (
-                    error_time + "_reference_" + str(uuid.uuid4()) + ".wav"
-                )
-                error_api = HfApi()
-                error_api.upload_file(
-                    path_or_fileobj=speaker_wav,
-                    path_in_repo=speaker_filename,
-                    repo_id="coqui/xtts-flagged-dataset",
-                    repo_type="dataset",
-                )
-
-                # HF Space specific.. This error is unrecoverable need to restart space
-                space = api.get_space_runtime(repo_id=repo_id)
-                if space.stage!="BUILDING":
-                    api.restart_space(repo_id=repo_id)
-                else:
-                    print("TRIED TO RESTART but space is building")
-                    
+                
+                # HF Space specific.. This error is unrecoverable need to restart space 
+                api.restart_space(repo_id=repo_id)
             else:
-                if "Failed to decode" in str(e):
-                    print("Speaker encoding error", str(e))
-                    gr.Warning(
-                        "It appears something wrong with reference, did you unmute your microphone?"
-                    )
-                else:
-                    print("RuntimeError: non device-side assert error:", str(e))
-                    gr.Warning("Something unexpected happened please retry again.")
-                return (
-                    None,
-                    None,
-                    None,
-                    None,
-                )
+                print("RuntimeError: non device-side assert error:", str(e))
+                raise e
         return (
             gr.make_waveform(
                 audio="output.wav",
@@ -401,45 +227,45 @@ def predict(
     else:
         gr.Warning("Please accept the Terms & Condition!")
         return (
-            None,
-            None,
-            None,
-            None,
-        )
+                None,
+                None,
+                None,
+                None,
+            ) 
 
 
 title = "Coqui🐸 XTTS"
 
 description = """
+<div>
+<a style="display:inline-block" href='https://github.com/coqui-ai/TTS'><img src='https://img.shields.io/github/stars/coqui-ai/TTS?style=social' /></a>
+<a style='display:inline-block' href='https://discord.gg/5eXr5seRrv'><img src='https://discord.com/api/guilds/1037326658807533628/widget.png?style=shield' /></a>
+<a href="https://huggingface.co/spaces/coqui/xtts?duplicate=true">
+<img style="margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>
+</div>
 
+<a href="https://huggingface.co/coqui/XTTS-v1">XTTS</a> is a Voice generation model that lets you clone voices into different languages by using just a quick 6-second audio clip. 
 <br/>
-
-This demo is currently running **XTTS v2.0.3** <a href="https://huggingface.co/coqui/XTTS-v2">XTTS</a> is a multilingual text-to-speech and voice-cloning model. This demo features zero-shot voice cloning, however, you can fine-tune XTTS for better results. Leave a star 🌟 on Github <a href="https://github.com/coqui-ai/TTS">🐸TTS</a>, where our open-source inference and training code lives.
-
+XTTS is built on previous research, like Tortoise, with additional architectural innovations and training to make cross-language voice cloning and multilingual speech generation possible. 
 <br/>
-
-Supported languages: Arabic: ar, Brazilian Portuguese: pt , Mandarin Chinese: zh-cn, Czech: cs, Dutch: nl, English: en, French: fr, German: de, Italian: it, Polish: pl, Russian: ru, Spanish: es, Turkish: tr, Japanese: ja, Korean: ko, Hungarian: hu, Hindi: hi
-
+This is the same model that powers our creator application <a href="https://coqui.ai">Coqui Studio</a> as well as the <a href="https://docs.coqui.ai">Coqui API</a>. In production we apply modifications to make low-latency streaming possible.
 <br/>
-"""
-
-links = """
-<img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=0d00920c-8cc9-4bf3-90f2-a615797e5f59" />
-
-|                                 |                                         |
-| ------------------------------- | --------------------------------------- |
-| 🐸💬 **CoquiTTS**                | <a style="display:inline-block" href='https://github.com/coqui-ai/TTS'><img src='https://img.shields.io/github/stars/coqui-ai/TTS?style=social' /></a>|
-| 💼 **Documentation**            | [ReadTheDocs](https://tts.readthedocs.io/en/latest/)
-| 👩‍💻 **Questions**                | [GitHub Discussions](https://github.com/coqui-ai/TTS/discussions) |
-| 🗯 **Community**         | [![Dicord](https://img.shields.io/discord/1037326658807533628?color=%239B59B6&label=chat%20on%20discord)](https://discord.gg/5eXr5seRrv)  |
-
-
+Leave a star on the Github <a href="https://github.com/coqui-ai/TTS">🐸TTS</a>, where our open-source inference and training code lives.
+<br/>
+<p>For faster inference without waiting in the queue, you should duplicate this space and upgrade to GPU via the settings.
+<br/>
+</p>
+<p>Language Selectors: 
+Arabic: ar, Brazilian Portuguese: pt , Chinese: zh-cn, Czech: cs,<br/> 
+Dutch: nl, English: en, French: fr, Italian: it, Polish: pl,<br/> 
+Russian: ru, Spanish: es, Turkish: tr <br/> 
+</p>
+<img referrerpolicy="no-referrer-when-downgrade" src="https://static.scarf.sh/a.png?x-pxid=8946ef36-c454-4a8e-a9c9-8a8dd735fabd" />
 """
 
 article = """
 <div style='margin:20px auto;'>
 <p>By using this demo you agree to the terms of the Coqui Public Model License at https://coqui.ai/cpml</p>
-<p>We collect data only for error cases for improvement.</p>
 </div>
 """
 examples = [
@@ -452,6 +278,7 @@ examples = [
         False,
         False,
         True,
+
     ],
     [
         "Lorsque j'avais six ans j'ai vu, une fois, une magnifique image",
@@ -508,7 +335,7 @@ examples = [
         "it",
         "examples/female.wav",
         None,
-        False,
+       False,
         False,
         False,
         True,
@@ -528,7 +355,7 @@ examples = [
         "ru",
         "examples/female.wav",
         None,
-        False,
+       False,
         False,
         False,
         True,
@@ -538,7 +365,7 @@ examples = [
         "nl",
         "examples/male.wav",
         None,
-        False,
+       False,
         False,
         False,
         True,
@@ -548,7 +375,7 @@ examples = [
         "cs",
         "examples/female.wav",
         None,
-        False,
+       False,
         False,
         False,
         True,
@@ -558,146 +385,80 @@ examples = [
         "zh-cn",
         "examples/female.wav",
         None,
+       False,
         False,
         False,
-        False,
-        True,
-    ],
-    [
-        "かつて 六歳のとき、素晴らしい絵を見ました",
-        "ja",
-        "examples/female.wav",
-        None,
-        False,
-        True,
-        False,
-        True,
-    ],
-    [
-        "한번은 내가 여섯 살이었을 때 멋진 그림을 보았습니다.",
-        "ko",
-        "examples/female.wav",
-        None,
-        False,
-        True,
-        False,
-        True,
-    ],
-        [
-        "Egyszer hat éves koromban láttam egy csodálatos képet",
-        "hu",
-        "examples/male.wav",
-        None,
-        False,
-        True,
-        False,
         True,
     ],
 ]
 
 
 
-with gr.Blocks(analytics_enabled=False) as demo:
-    with gr.Row():
-        with gr.Column():
-            gr.Markdown(
-                """
-                ## <img src="https://raw.githubusercontent.com/coqui-ai/TTS/main/images/coqui-log-green-TTS.png" height="56"/>
-                """
-            )
-        with gr.Column():
-            # placeholder to align the image
-            pass
-
-    with gr.Row():
-        with gr.Column():
-            gr.Markdown(description)
-        with gr.Column():
-            gr.Markdown(links)
-
-    with gr.Row():
-        with gr.Column():
-            input_text_gr = gr.Textbox(
-                label="Text Prompt",
-                info="One or two sentences at a time is better. Up to 200 text characters.",
-                value="Hi there, I'm your new voice clone. Try your best to upload quality audio.",
-            )
-            language_gr = gr.Dropdown(
-                label="Language",
-                info="Select an output language for the synthesised speech",
-                choices=[
-                    "en",
-                    "es",
-                    "fr",
-                    "de",
-                    "it",
-                    "pt",
-                    "pl",
-                    "tr",
-                    "ru",
-                    "nl",
-                    "cs",
-                    "ar",
-                    "zh-cn",
-                    "ja",
-                    "ko",
-                    "hu",
-                    "hi"
-                ],
-                max_choices=1,
-                value="en",
-            )
-            ref_gr = gr.Audio(
-                label="Reference Audio",
-                info="Click on the ✎ button to upload your own target speaker audio",
-                type="filepath",
-                value="examples/female.wav",
-            )
-            mic_gr = gr.Audio(
-                source="microphone",
-                type="filepath",
-                info="Use your microphone to record audio",
-                label="Use Microphone for Reference",
-            )
-            use_mic_gr = gr.Checkbox(
-                label="Use Microphone",
-                value=False,
-                info="Notice: Microphone input may not work properly under traffic",
-            )
-            clean_ref_gr = gr.Checkbox(
-                label="Cleanup Reference Voice",
-                value=False,
-                info="This check can improve output if your microphone or reference voice is noisy",
-            )
-            auto_det_lang_gr = gr.Checkbox(
-                label="Do not use language auto-detect",
-                value=False,
-                info="Check to disable language auto-detection",
-            )
-            tos_gr = gr.Checkbox(
-                label="Agree",
-                value=False,
-                info="I agree to the terms of the CPML: https://coqui.ai/cpml",
-            )
-
-            tts_button = gr.Button("Send", elem_id="send-btn", visible=True)
-
-
-        with gr.Column():
-            video_gr = gr.Video(label="Waveform Visual")
-            audio_gr = gr.Audio(label="Synthesised Audio", autoplay=True)
-            out_text_gr = gr.Text(label="Metrics")
-            ref_audio_gr = gr.Audio(label="Reference Audio Used")
-
-    with gr.Row():
-        gr.Examples(examples,
-                    label="Examples",
-                    inputs=[input_text_gr, language_gr, ref_gr, mic_gr, use_mic_gr, clean_ref_gr, auto_det_lang_gr, tos_gr],
-                    outputs=[video_gr, audio_gr, out_text_gr, ref_audio_gr],
-                    fn=predict,
-                    cache_examples=False,)
-
-    tts_button.click(predict, [input_text_gr, language_gr, ref_gr, mic_gr, use_mic_gr, clean_ref_gr, auto_det_lang_gr, tos_gr], outputs=[video_gr, audio_gr, out_text_gr, ref_audio_gr])
-
-demo.queue()  
-demo.launch(debug=True, show_api=True)
+gr.Interface(
+    fn=predict,
+    inputs=[
+        gr.Textbox(
+            label="Text Prompt",
+            info="One or two sentences at a time is better. Up to 200 text characters.",
+            value="Hi there, I'm your new voice clone. Try your best to upload quality audio",
+        ),
+        gr.Dropdown(
+            label="Language",
+            info="Select an output language for the synthesised speech",
+            choices=[
+                "en",
+                "es",
+                "fr",
+                "de",
+                "it",
+                "pt",
+                "pl",
+                "tr",
+                "ru",
+                "nl",
+                "cs",
+                "ar",
+                "zh-cn",
+            ],
+            max_choices=1,
+            value="en",
+        ),
+        gr.Audio(
+            label="Reference Audio",
+            info="Click on the ✎ button to upload your own target speaker audio",
+            type="filepath",
+            value="examples/female.wav",
+        ),
+        gr.Audio(source="microphone",
+                 type="filepath",
+                 info="Use your microphone to record audio",
+                 label="Use Microphone for Reference"),
+        gr.Checkbox(label="Use Microphone",
+                    value=False,
+                    info="Notice: Microphone input may not work properly under traffic",),
+        gr.Checkbox(label="Cleanup Reference Voice",
+                    value=False,
+                    info="This check can improve output if your microphone or reference voice is noisy",
+                    ),
+        gr.Checkbox(label="Do not use language auto-detect",
+                    value=False,
+                    info="Check to disable language auto-detection",),
+        gr.Checkbox(
+            label="Agree",
+            value=False,
+            info="I agree to the terms of the Coqui Public Model License at https://coqui.ai/cpml",
+        ),
+
+        
+    ],
+    outputs=[
+        gr.Video(label="Waveform Visual"),
+        gr.Audio(label="Synthesised Audio",autoplay=True),
+        gr.Text(label="Metrics"),
+        gr.Audio(label="Reference Audio Used"),
+    ],
+    title=title,
+    description=description,
+    article=article,
+    examples=examples,
+).queue().launch(debug=True,show_api=False)

requirements.txt

@@ -1,12 +1,59 @@
 # Preinstall requirements from TTS 
-TTS @ git+https://github.com/coqui-ai/TTS@v0.21.1
-pydantic==1.10.13
-python-multipart==0.0.6
-typing-extensions>=4.8.0
-cutlet
-mecab-python3==1.0.6
-unidic-lite==1.0.8
-unidic==1.1.0
+torch==2.0.1 --index-url https://download.pytorch.org/whl/cu118
+torchvision==0.15.2 --index-url https://download.pytorch.org/whl/cu118 
+torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
+numpy==1.22.0;python_version<="3.10"
+numpy==1.24.3;python_version>"3.10"
+cython==0.29.30
+scipy>=1.11.2
+soundfile==0.12.*
+librosa==0.10.*
+scikit-learn==1.3.0
+numba==0.55.1;python_version<"3.9"
+numba==0.57.0;python_version>="3.9"
+inflect==5.6.*
+tqdm==4.64.*
+anyascii==0.3.*
+pyyaml==6.*
+fsspec==2023.6.0 # <= 2023.9.1 makes aux tests fail
+aiohttp==3.8.*
+packaging==23.1
+# deps for examples
+flask==2.*
+# deps for inference
+pysbd==0.3.4
+# deps for notebooks
+umap-learn==0.5.*
+pandas>=1.4,<2.0
+# deps for training
+matplotlib==3.7.*
+# coqui stack
+trainer
+# config management
+coqpit>=0.0.16
+# chinese g2p deps
+jieba
+pypinyin==0.47.1
+# gruut+supported langs
+gruut[de,es,fr]==2.2.3
+# deps for korean
+jamo
+nltk
+g2pkk>=0.1.1
+# deps for bangla
+bangla
+bnnumerizer
+bnunicodenormalizer
+#deps for tortoise
+k_diffusion
+einops==0.6.*
+transformers==4.33.*
+#deps for bark
+encodec==0.1.*
+# deps for XTTS
+unidecode==1.3.*
 langid
-deepspeed
-pydub
+# Install tts
+git+https://github.com/coqui-ai/tts.git@v0.17.7
+deepspeed==0.8.3
+pydub