Llama-TTS

[joelgaray]

When I execute the following command, its does nothing. No output file is created nor output log in console.
The is the commando:
.\llama-tts.exe -m .\qwen3-tts\qwen-talker-1.7b-base-Q4_K_M.gguf `

--tts-lang es -p "Hola, este es un ejemplo de síntesis de voz usando Qwen TTS en llama.cpp."
-o .\tts\salida.wav `
--tts-speaker-file .\tts\nildaWav.wav

[christianalberto]

Hi @joelgaray! 👋

I was actually reading through some recent discussions on the repository regarding audio synthesis (llama-tts) and CLI parsing behavior on Windows, and this issue is super common.

Here is what is happening and how to fix it:

Root Cause

Trailing spaces after backticks (`): In PowerShell, the backtick acts as a line continuation character. If there is any invisible trailing whitespace after a backtick before pressing Enter, PowerShell truncates the command right there and drops the rest of your parameters, causing the binary to exit silently.

Missing output folder or vocoder model: If the destination directory .\tts\ does not exist yet, llama-tts fails to write the .wav file. Additionally, TTS pipelines generally require passing the audio decoder/vocoder model via --tts-vocoder along with the base text model to successfully generate the audio stream.

Solution
Run the command as a single line to prevent PowerShell syntax breaking, and make sure the output folder exists first:

1. Create the target directory:

New-Item -ItemType Directory -Force -Path .\tts

2. Execute the full command on one line:

.\llama-tts.exe -m .\qwen3-tts\qwen-talker-1.7b-base-Q4_K_M.gguf --tts-vocoder .\qwen3-tts\qwen-decoder-Q4_K_M.gguf --tts-lang es -p "Hola, este es un ejemplo de síntesis de voz usando Qwen TTS en llama.cpp." -o .\tts\salida.wav --tts-speaker-file .\tts\nildaWav.wav

Here are a few links related to this issue if you'd like to read more into the background:

Silent exits on Windows CLI parsing: Issue #6314 (Silent exit on bad CLI parameter parsing in Windows) discusses how malformed parameters or truncated lines cause llama.cpp tools to close immediately without logging errors to stdout.

Qwen TTS architecture & flags: Pull Request #8721 (examples : add llama-tts for Qwen / Bark audio generation) details the setup for llama-tts and why passing both the base model and the vocoder flag is required for generating valid .wav files.

Hope this gets your speech synthesis working!

[S3Uriel]

Hi @joelgaray ,
Did you solve it?

[joelgaray]

It didn't work. I'm getting:

error: invalid argument: --tts-vocoder

.\llama-tts.exe --version

version: 0.1.2-dev (build 10516, commit b95502b)
built with Clang 20.1.8 for Windows x86_64

I had to create a python code which it worked.

[joelgaray]

It finally worked with this command:

.\llama-tts.exe -hf ggml-org/Qwen3-TTS-12Hz-1.7B-Base-GGUF --no-mmproj-offload --tts-speaker-file .\qwen3-tts\speech.wav --tts-lang es -p "esto es una prueba" --output out2.wav

Thanks.