Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/IAHispano/Applio/llms.txt

Use this file to discover all available pages before exploring further.

Applio’s TTS feature combines two distinct systems to give you fully synthetic speech in any trained voice: first, Microsoft Edge TTS (via the edge-tts library) converts your text to natural-sounding speech using one of hundreds of neural voices, then the output WAV is immediately passed through the standard RVC inference pipeline to re-voice it using your chosen .pth model. The final result is an audio file that sounds like the target model speaker saying your text. Both the intermediate TTS audio and the final RVC-converted audio are saved separately, so you can inspect or use either one.
The TTS voice list is fetched from the Bing TTS endpoint (https://speech.platform.bing.com/consumer/speech/synthesize/readaloud/voices/list) and cached locally in rvc/lib/tools/tts_voices.json. This file ships with Applio and is loaded at startup via load_voices_data() in core.py. The ShortName field of each voice entry (e.g. en-US-AriaNeural) is what you pass as tts_voice.

​
How the Pipeline Works

1

Text to Speech (Edge TTS)

Applio calls rvc/lib/tools/tts.py as a subprocess, passing your text (or a UTF-8 text file), the selected Edge TTS voice short name, and a speech rate adjustment. Edge TTS synthesises a WAV file and saves it to output_tts_path (default: assets/audios/tts_output.wav). If a file already exists at that path it is deleted first.
2

Voice Conversion (RVC)

The synthesised WAV is fed directly into VoiceConverter.convert_audio() — exactly the same function used by single-file inference. The converted audio is saved to output_rvc_path (default: assets/audios/tts_rvc_output.wav).

​
Parameters

​
tts_text
str
The text string to synthesise. Used when you type directly in the “Text to Speech” tab. Mutually exclusive with tts_file — if both are provided, tts_file takes precedence when a valid path is given.
​
tts_file
str
Path to a UTF-8 encoded .txt file containing the text to synthesise. The full file content is passed to Edge TTS. The file must be UTF-8; other encodings will cause an error.
​
tts_voice
str
required
The Edge TTS voice short name to use for synthesis (e.g. en-US-AriaNeural, es-ES-AlvaroNeural, ja-JP-NanamiNeural). The complete list of available voice short names is loaded from rvc/lib/tools/tts_voices.json at startup.
​
tts_rate
int
default:"0"
Adjusts the Edge TTS speech rate as a percentage relative to the voice’s normal speed. The range is -100 to +100. A value of 0 uses the default speed; -50 speaks at half speed; +50 speeds up by 50%.
​
output_tts_path
str
default:"assets/audios/tts_output.wav"
Path where the intermediate Edge TTS audio is saved before RVC conversion.
​
output_rvc_path
str
default:"assets/audios/tts_rvc_output.wav"
Path where the final RVC-converted audio is saved.
​
pth_path
str
required
Path to the .pth model file used for voice conversion.
​
index_path
str
required
Path to the .index file paired with the model.
​
pitch
int
default:"0"
Pitch shift in semitones applied during the RVC conversion step (-24 to +24).
​
f0_method
str
default:"rmvpe"
Pitch extraction algorithm for the RVC conversion step. Choices: crepe, crepe-tiny, rmvpe, fcpe.
​
index_rate
float
default:"0.75"
Index file influence during RVC conversion (0.0–1.0). The TTS tab defaults to 0.75 (higher than the standard inference default of 0.3) because synthesised speech is clean and benefits from stronger index guidance.
​
volume_envelope
float
default:"1.0"
Volume envelope blending for the RVC output (0.0–1.0).
​
protect
float
default:"0.5"
Consonant and breath protection level for RVC conversion (0.0–0.5). The TTS tab defaults to 0.5 (maximum protection) since synthesised TTS speech is already clean.
​
split_audio
bool
default:"false"
Whether to split the TTS output into chunks before RVC conversion. Useful for long texts.
​
f0_autotune
bool
default:"false"
Apply autotune to the RVC output. Useful when converting song lyrics via TTS.
​
clean_audio
bool
default:"false"
Apply noise reduction to the final RVC output.
​
clean_strength
float
default:"0.5"
Noise reduction intensity (0.0–1.0).
​
export_format
str
default:"WAV"
Output audio format. Choices: WAV, MP3, FLAC, OGG, M4A.
​
embedder_model
str
default:"contentvec"
Speaker-embedding model for the RVC conversion step. Must match the embedder used when training the model. Choices: contentvec, spin, spin-v2, chinese-hubert-base, japanese-hubert-base, korean-hubert-base, custom.
​
sid
int
default:"0"
Speaker ID for multi-speaker models.

​
CLI Example

python core.py tts \
  --tts_text "Hello, this is a test." \
  --tts_voice en-US-AriaNeural \
  --tts_rate 0 \
  --pth_path logs/MyModel/MyModel.pth \
  --index_path logs/MyModel/MyModel.index \
  --output_tts_path assets/audios/tts_output.wav \
  --output_rvc_path assets/audios/rvc_output.wav \
  --pitch 0 \
  --f0_method rmvpe \
  --index_rate 0.75 \
  --volume_envelope 1.0 \
  --protect 0.5 \
  --export_format WAV \
  --embedder_model contentvec \
  --clean_audio False
To use a text file instead of inline text: 
python core.py tts \
  --tts_file path/to/my_script.txt \
  --tts_voice en-GB-SoniaNeural \
  --tts_rate -10 \
  --pth_path logs/MyModel/MyModel.pth \
  --index_path logs/MyModel/MyModel.index \
  --output_tts_path assets/audios/tts_output.wav \
  --output_rvc_path assets/audios/rvc_output.wav

​
Python API

from core import run_tts_script

message, output_path = run_tts_script(
    tts_file="",                    # set to a file path, or leave empty to use tts_text
    tts_text="Hello, this is a test.",
    tts_voice="en-US-AriaNeural",
    tts_rate=0,
    pitch=0,
    index_rate=0.75,
    volume_envelope=1.0,
    protect=0.5,
    f0_method="rmvpe",
    output_tts_path="assets/audios/tts_output.wav",
    output_rvc_path="assets/audios/rvc_output.wav",
    pth_path="logs/MyModel/MyModel.pth",
    index_path="logs/MyModel/MyModel.index",
    split_audio=False,
    f0_autotune=False,
    f0_autotune_strength=1.0,
    proposed_pitch=False,
    proposed_pitch_threshold=155.0,
    clean_audio=False,
    clean_strength=0.5,
    export_format="WAV",
    embedder_model="contentvec",
    embedder_model_custom=None,
    sid=0,
)

print(message)      # "Text Hello, this is a test. synthesized successfully."
print(output_path)  # "assets/audios/rvc_output.wav"
You can find all available voice short names by inspecting the ShortName field in rvc/lib/tools/tts_voices.json. The file contains hundreds of voices across dozens of languages and locales. Pass any ShortName value directly to --tts_voice or the tts_voice parameter.
Text files passed via tts_file must be UTF-8 encoded. Non-UTF-8 files will fail to load and the conversion will not start. The Gradio UI will display a warning if encoding detection fails.

Build docs developers (and LLMs) love

Get started for freeTalk to us