piper-tts 命令

常用示例

说明

piper is an offline neural text-to-speech engine that runs VITS voice models exported to ONNX. Installing the piper-tts Python package provides the piper binary, which reads text from standard input (or the file given by --input-file) and writes 16-bit PCM WAV audio either to a file or to standard output for streaming. Voices are distributed separately and downloaded with python3 -m piper.download_voices. Models cover many languages and accents, and several are multi-speaker — use --speaker to pick a voice. Phonemization is performed via embedded espeak-ng, and inline overrides written in double brackets (for example `[ bˈætmæn ]`) are accepted for fine-grained pronunciation control. Output quality and latency depend on the model variant (_x_low_, _low_, _medium_, _high_). For interactive use, piper is typically wrapped behind a daemon so the model is loaded once instead of on every invocation.

参数

-m, --model _voice_

Voice model identifier (e.g. _en_US-lessac-medium_) or path to an ONNX file.

-f, --output-file _file_

Write the synthesized WAV audio to _file_.

--output-raw

Write raw 16-bit PCM audio to stdout (for piping into a player).

--input-file _file_

Read input text from _file_ instead of stdin.

--data-dir _dir_

Directory where downloaded voice files are stored.

--speaker _n_

Select speaker index for multi-speaker voices.

--sentence-silence _seconds_

Seconds of silence inserted between sentences.

--volume _factor_

Output volume multiplier (1.0 is unchanged).

--cuda

Use the CUDA execution provider (requires onnxruntime-gpu).

--json-input

Read JSON objects from stdin instead of plain text.