This project extends the original talk-llama demo to seamlessly combine Whisper speech recognition with local LLMs served by LM Studio.
You speak → Whisper transcribes → a model loaded in LM Studio generates the reply → optional TTS speaks it.

Sorry for not being a coder, i did this with the help of ChatGPT 5, while having a nice Sunday afternoon :-)

Jump to “LM Studio backend & new options” below for setup, CLI flags, build notes, and troubleshooting.

Stable: v1.7.6 / Roadmap

High-performance inference of OpenAI's Whisper automatic speech recognition (ASR) model:

• Plain C/C++ implementation without dependencies
• Apple Silicon first-class citizen - optimized via ARM NEON, Accelerate framework, Metal and Core ML
• AVX intrinsics support for x86 architectures
• VSX intrinsics support for POWER architectures
• Mixed F16 / F32 precision
• Integer quantization support
• Zero memory allocations at runtime
• Vulkan support
• Support for CPU-only inference
• Efficient GPU support for NVIDIA
• OpenVINO Support
• Ascend NPU Support
• Moore Threads GPU Support
• C-style API
• Voice Activity Detection (VAD)

Supported platforms:

• Mac OS (Intel and Arm)
• iOS
• Android
• Java
• Linux / FreeBSD
• WebAssembly
• Windows (MSVC and MinGW)
• Raspberry Pi
• Docker

The entire high-level implementation of the model is contained in whisper.h and whisper.cpp. The rest of the code is part of the ggml machine learning library.

Having such a lightweight implementation of the model allows to easily integrate it in different platforms and applications. As an example, here is a video of running the model on an iPhone 13 device - fully offline, on-device: whisper.objc

You can also easily make your own offline voice assistant application: command

On Apple Silicon, the inference runs fully on the GPU via Metal:

First clone the repository:

git clone https://github.com/ggml-org/whisper.cpp.git

Navigate into the directory:

cd whisper.cpp

Then, download one of the Whisper models converted in ggml format. For example:

sh ./models/download-ggml-model.sh base.en

Now build the whisper-cli example and transcribe an audio file like this:

# build the project
cmake -B build
cmake --build build -j --config Release

# transcribe an audio file
./build/bin/whisper-cli -f samples/jfk.wav

For a quick demo, simply run make base.en.

The command downloads the base.en model converted to custom ggml format and runs the inference on all .wav samples in the folder samples.

For detailed usage instructions, run: ./build/bin/whisper-cli -h

Note that the whisper-cli example currently runs only with 16-bit WAV files, so make sure to convert your input before running the tool. For example, you can use ffmpeg like this:

ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav

If you want some extra audio samples to play with, simply run:

make -j samples

This will download a few more audio files from Wikipedia and convert them to 16-bit WAV format via ffmpeg.

You can download and run the other models as follows:

make -j tiny.en
make -j tiny
make -j base.en
make -j base
make -j small.en
make -j small
make -j medium.en
make -j medium
make -j large-v1
make -j large-v2
make -j large-v3
make -j large-v3-turbo

whisper.cpp supports POWER architectures and includes code which significantly speeds operation on Linux running on POWER9/10, making it capable of faster-than-realtime transcription on underclocked Raptor Talos II. Ensure you have a BLAS package installed, and replace the standard cmake setup with:

# build with GGML_BLAS defined
cmake -B build -DGGML_BLAS=1
cmake --build build -j --config Release
./build/bin/whisper-cli [ .. etc .. ]

whisper.cpp supports integer quantization of the Whisper ggml models. Quantized models require less memory and disk space and depending on the hardware can be processed more efficiently.

Here are the steps for creating and using a quantized model:

# quantize a model with Q5_0 method
cmake -B build
cmake --build build -j --config Release
./build/bin/quantize models/ggml-base.en.bin models/ggml-base.en-q5_0.bin q5_0

# run the examples as usual, specifying the quantized model file
./build/bin/whisper-cli -m models/ggml-base.en-q5_0.bin ./samples/gb0.wav

On Apple Silicon devices, the Encoder inference can be executed on the Apple Neural Engine (ANE) via Core ML. This can result in significant speed-up - more than x3 faster compared with CPU-only execution. Here are the instructions for generating a Core ML model and using it with whisper.cpp:

• Install Python dependencies needed for the creation of the Core ML model:

  pip install ane_transformers
  pip install openai-whisper
  pip install coremltools

  • To ensure coremltools operates correctly, please confirm that Xcode is installed and execute xcode-select --install to install the command-line tools.
  • Python 3.11 is recommended.
  • MacOS Sonoma (version 14) or newer is recommended, as older versions of MacOS might experience issues with transcription hallucination.
  • [OPTIONAL] It is recommended to utilize a Python version management system, such as Miniconda for this step:
    • To create an environment, use: conda create -n py311-whisper python=3.11 -y
    • To activate the environment, use: conda activate py311-whisper

• Generate a Core ML model. For example, to generate a base.en model, use:

  ./models/generate-coreml-model.sh base.en

  This will generate the folder models/ggml-base.en-encoder.mlmodelc

• Build whisper.cpp with Core ML support:

  # using CMake
  cmake -B build -DWHISPER_COREML=1
  cmake --build build -j --config Release

• Run the examples as usual. For example:

  $ ./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/jfk.wav
  
  ...
  
  whisper_init_state: loading Core ML model from 'models/ggml-base.en-encoder.mlmodelc'
  whisper_init_state: first run on a device may take a while ...
  whisper_init_state: Core ML model loaded
  
  system_info: n_threads = 4 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 | COREML = 1 |
  
  ...
  

  The first run on a device is slow, since the ANE service compiles the Core ML model to some device-specific format. Next runs are faster.

For more information about the Core ML implementation please refer to PR #566.

On platforms that support OpenVINO, the Encoder inference can be executed on OpenVINO-supported devices including x86 CPUs and Intel GPUs (integrated & discrete).

This can result in significant speedup in encoder performance. Here are the instructions for generating the OpenVINO model and using it with whisper.cpp:

• First, setup python virtual env. and install python dependencies. Python 3.10 is recommended.

  Windows:

  cd models
  python -m venv openvino_conv_env
  openvino_conv_env\Scripts\activate
  python -m pip install --upgrade pip
  pip install -r requirements-openvino.txt

  Linux and macOS:

  cd models
  python3 -m venv openvino_conv_env
  source openvino_conv_env/bin/activate
  python -m pip install --upgrade pip
  pip install -r requirements-openvino.txt

• Generate an OpenVINO encoder model. For example, to generate a base.en model, use:

  python convert-whisper-to-openvino.py --model base.en
  

  This will produce ggml-base.en-encoder-openvino.xml/.bin IR model files. It's recommended to relocate these to the same folder as ggml models, as that is the default location that the OpenVINO extension will search at runtime.

• Build whisper.cpp with OpenVINO support:

  Download OpenVINO package from release page. The recommended version to use is 2024.6.0. Ready to use Binaries of the required libraries can be found in the OpenVino Archives

  After downloading & extracting package onto your development system, set up required environment by sourcing setupvars script. For example:

  Linux:

  source /path/to/l_openvino_toolkit_ubuntu22_2023.0.0.10926.b4452d56304_x86_64/setupvars.sh

  Windows (cmd):

  C:\Path\To\w_openvino_toolkit_windows_2023.0.0.10926.b4452d56304_x86_64\setupvars.bat

  And then build the project using cmake:

  cmake -B build -DWHISPER_OPENVINO=1
  cmake --build build -j --config Release

• Run the examples as usual. For example:

  $ ./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/jfk.wav
  
  ...
  
  whisper_ctx_init_openvino_encoder: loading OpenVINO model from 'models/ggml-base.en-encoder-openvino.xml'
  whisper_ctx_init_openvino_encoder: first run on a device may take a while ...
  whisper_openvino_init: path_model = models/ggml-base.en-encoder-openvino.xml, device = GPU, cache_dir = models/ggml-base.en-encoder-openvino-cache
  whisper_ctx_init_openvino_encoder: OpenVINO model loaded
  
  system_info: n_threads = 4 / 8 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | VSX = 0 | COREML = 0 | OPENVINO = 1 |
  
  ...
  

  The first time run on an OpenVINO device is slow, since the OpenVINO framework will compile the IR (Intermediate Representation) model to a device-specific 'blob'. This device-specific blob will get cached for the next run.

For more information about the OpenVINO implementation please refer to PR #1037.

With NVIDIA cards the processing of the models is done efficiently on the GPU via cuBLAS and custom CUDA kernels. First, make sure you have installed cuda: https://developer.nvidia.com/cuda-downloads

Now build whisper.cpp with CUDA support:

cmake -B build -DGGML_CUDA=1
cmake --build build -j --config Release

or for newer NVIDIA GPU's (RTX 5000 series):

cmake -B build -DGGML_CUDA=1 -DCMAKE_CUDA_ARCHITECTURES="86"
cmake --build build -j --config Release

Cross-vendor solution which allows you to accelerate workload on your GPU. First, make sure your graphics card driver provides support for Vulkan API.

Now build whisper.cpp with Vulkan support:

cmake -B build -DGGML_VULKAN=1
cmake --build build -j --config Release

Encoder processing can be accelerated on the CPU via OpenBLAS. First, make sure you have installed openblas: https://www.openblas.net/

Now build whisper.cpp with OpenBLAS support:

cmake -B build -DGGML_BLAS=1
cmake --build build -j --config Release

Ascend NPU provides inference acceleration via CANN and AI cores.

First, check if your Ascend NPU device is supported:

Verified devices

Then, make sure you have installed CANN toolkit . The lasted version of CANN is recommanded.

Now build whisper.cpp with CANN support:

cmake -B build -DGGML_CANN=1
cmake --build build -j --config Release

Run the inference examples as usual, for example:

./build/bin/whisper-cli -f samples/jfk.wav -m models/ggml-base.en.bin -t 8

Notes:

• If you have trouble with Ascend NPU device, please create a issue with [CANN] prefix/tag.
• If you run successfully with your Ascend NPU device, please help update the table Verified devices.

With Moore Threads cards the processing of the models is done efficiently on the GPU via muBLAS and custom MUSA kernels. First, make sure you have installed MUSA SDK rc4.2.0: https://developer.mthreads.com/sdk/download/musa?equipment=&os=&driverVersion=&version=4.2.0

Now build whisper.cpp with MUSA support:

cmake -B build -DGGML_MUSA=1
cmake --build build -j --config Release

or specify the architecture for your Moore Threads GPU. For example, if you have a MTT S80 GPU, you can specify the architecture as follows:

cmake -B build -DGGML_MUSA=1 -DMUSA_ARCHITECTURES="21"
cmake --build build -j --config Release

If you want to support more audio formats (such as Opus and AAC), you can turn on the WHISPER_FFMPEG build flag to enable FFmpeg integration.

First, you need to install required libraries:

# Debian/Ubuntu
sudo apt install libavcodec-dev libavformat-dev libavutil-dev

# RHEL/Fedora
sudo dnf install libavcodec-free-devel libavformat-free-devel libavutil-free-devel

Then you can build the project as follows:

cmake -B build -D WHISPER_FFMPEG=yes
cmake --build build

Run the following example to confirm it's working:

# Convert an audio file to Opus format
ffmpeg -i samples/jfk.wav jfk.opus

# Transcribe the audio file
./build/bin/whisper-cli --model models/ggml-base.en.bin --file jfk.opus

• Docker must be installed and running on your system.
• Create a folder to store big models & intermediate files (ex. /whisper/models)

We have two Docker images available for this project:

1. ghcr.io/ggml-org/whisper.cpp:main: This image includes the main executable file as well as curl and ffmpeg. (platforms: linux/amd64, linux/arm64)
2. ghcr.io/ggml-org/whisper.cpp:main-cuda: Same as main but compiled with CUDA support. (platforms: linux/amd64)
3. ghcr.io/ggml-org/whisper.cpp:main-musa: Same as main but compiled with MUSA support. (platforms: linux/amd64)

# download model and persist it in a local folder
docker run -it --rm \
  -v path/to/models:/models \
  whisper.cpp:main "./models/download-ggml-model.sh base /models"
# transcribe an audio file
docker run -it --rm \
  -v path/to/models:/models \
  -v path/to/audios:/audios \
  whisper.cpp:main "whisper-cli -m /models/ggml-base.bin -f /audios/jfk.wav"
# transcribe an audio file in samples folder
docker run -it --rm \
  -v path/to/models:/models \
  whisper.cpp:main "whisper-cli -m /models/ggml-base.bin -f ./samples/jfk.wav"

You can install pre-built binaries for whisper.cpp or build it from source using Conan. Use the following command:

conan install --requires="whisper-cpp/[*]" --build=missing

For detailed instructions on how to use Conan, please refer to the Conan documentation.

• Inference only

This is a naive example of performing real-time inference on audio from your microphone. The stream tool samples the audio every half a second and runs the transcription continuously. More info is available in issue #10. You will need to have sdl2 installed for it to work properly.

cmake -B build -DWHISPER_SDL2=ON
cmake --build build -j --config Release
./build/bin/whisper-stream -m ./models/ggml-base.en.bin -t 8 --step 500 --length 5000

Adding the --print-colors argument will print the transcribed text using an experimental color coding strategy to highlight words with high or low confidence:

./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/gb0.wav --print-colors

For example, to limit the line length to a maximum of 16 characters, simply add -ml 16:

$ ./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16

whisper_model_load: loading model from './models/ggml-base.en.bin'
...
system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |

main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...

[00:00:00.000 --> 00:00:00.850]   And so my
[00:00:00.850 --> 00:00:01.590]   fellow
[00:00:01.590 --> 00:00:04.140]   Americans, ask
[00:00:04.140 --> 00:00:05.660]   not what your
[00:00:05.660 --> 00:00:06.840]   country can do
[00:00:06.840 --> 00:00:08.430]   for you, ask
[00:00:08.430 --> 00:00:09.440]   what you can do
[00:00:09.440 --> 00:00:10.020]   for your
[00:00:10.020 --> 00:00:11.000]   country.

The --max-len argument can be used to obtain word-level timestamps. Simply use -ml 1:

$ ./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 1

whisper_model_load: loading model from './models/ggml-base.en.bin'
...
system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |

main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...

[00:00:00.000 --> 00:00:00.320]
[00:00:00.320 --> 00:00:00.370]   And
[00:00:00.370 --> 00:00:00.690]   so
[00:00:00.690 --> 00:00:00.850]   my
[00:00:00.850 --> 00:00:01.590]   fellow
[00:00:01.590 --> 00:00:02.850]   Americans
[00:00:02.850 --> 00:00:03.300]  ,
[00:00:03.300 --> 00:00:04.140]   ask
[00:00:04.140 --> 00:00:04.990]   not
[00:00:04.990 --> 00:00:05.410]   what
[00:00:05.410 --> 00:00:05.660]   your
[00:00:05.660 --> 00:00:06.260]   country
[00:00:06.260 --> 00:00:06.600]   can
[00:00:06.600 --> 00:00:06.840]   do
[00:00:06.840 --> 00:00:07.010]   for
[00:00:07.010 --> 00:00:08.170]   you
[00:00:08.170 --> 00:00:08.190]  ,
[00:00:08.190 --> 00:00:08.430]   ask
[00:00:08.430 --> 00:00:08.910]   what
[00:00:08.910 --> 00:00:09.040]   you
[00:00:09.040 --> 00:00:09.320]   can
[00:00:09.320 --> 00:00:09.440]   do
[00:00:09.440 --> 00:00:09.760]   for
[00:00:09.760 --> 00:00:10.020]   your
[00:00:10.020 --> 00:00:10.510]   country
[00:00:10.510 --> 00:00:11.000]  .

More information about this approach is available here: ggml-org#1058

Sample usage:

# download a tinydiarize compatible model
./models/download-ggml-model.sh small.en-tdrz

# run as usual, adding the "-tdrz" command-line argument
./build/bin/whisper-cli -f ./samples/a13.wav -m ./models/ggml-small.en-tdrz.bin -tdrz
...
main: processing './samples/a13.wav' (480000 samples, 30.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, tdrz = 1, timestamps = 1 ...
...
[00:00:00.000 --> 00:00:03.800]   Okay Houston, we've had a problem here. [SPEAKER_TURN]
[00:00:03.800 --> 00:00:06.200]   This is Houston. Say again please. [SPEAKER_TURN]
[00:00:06.200 --> 00:00:08.260]   Uh Houston we've had a problem.
[00:00:08.260 --> 00:00:11.320]   We've had a main beam up on a volt. [SPEAKER_TURN]
[00:00:11.320 --> 00:00:13.820]   Roger main beam interval. [SPEAKER_TURN]
[00:00:13.820 --> 00:00:15.100]   Uh uh [SPEAKER_TURN]
[00:00:15.100 --> 00:00:18.020]   So okay stand, by thirteen we're looking at it. [SPEAKER_TURN]
[00:00:18.020 --> 00:00:25.740]   Okay uh right now uh Houston the uh voltage is uh is looking good um.
[00:00:27.620 --> 00:00:29.940]   And we had a a pretty large bank or so.

The whisper-cli example provides support for output of karaoke-style movies, where the currently pronounced word is highlighted. Use the -owts argument and run the generated bash script. This requires to have ffmpeg installed.

Here are a few "typical" examples:

./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -owts
source ./samples/jfk.wav.wts
ffplay ./samples/jfk.wav.mp4

./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/mm0.wav -owts
source ./samples/mm0.wav.wts
ffplay ./samples/mm0.wav.mp4

./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/gb0.wav -owts
source ./samples/gb0.wav.wts
ffplay ./samples/gb0.wav.mp4

Use the scripts/bench-wts.sh script to generate a video in the following format:

./scripts/bench-wts.sh samples/jfk.wav
ffplay ./samples/jfk.wav.all.mp4

In order to have an objective comparison of the performance of the inference across different system configurations, use the whisper-bench tool. The tool simply runs the Encoder part of the model and prints how much time it took to execute it. The results are summarized in the following Github issue:

Benchmark results

Additionally a script to run whisper.cpp with different models and audio files is provided bench.py.

You can run it with the following command, by default it will run against any standard model in the models folder.

python3 scripts/bench.py -f samples/jfk.wav -t 2,4,8 -p 1,2

It is written in python with the intention of being easy to modify and extend for your benchmarking use case.

It outputs a csv file with the results of the benchmarking.

The original models are converted to a custom binary format. This allows to pack everything needed into a single file:

• model parameters
• mel filters
• vocabulary
• weights

You can download the converted models using the models/download-ggml-model.sh script or manually from here:

• https://huggingface.co/ggerganov/whisper.cpp

For more details, see the conversion script models/convert-pt-to-ggml.py or models/README.md.

• Rust: tazz4843/whisper-rs | #310
• JavaScript: bindings/javascript | #309
  • React Native (iOS / Android): whisper.rn
• Go: bindings/go | #312
• Java:
  • GiviMAD/whisper-jni
• Ruby: bindings/ruby | #507
• Objective-C / Swift: ggml-org/whisper.spm | #313
  • exPHAT/SwiftWhisper
• .NET: | #422
  • sandrohanea/whisper.net
  • NickDarvey/whisper
• Python: | #9
  • stlukey/whispercpp.py (Cython)
  • AIWintermuteAI/whispercpp (Updated fork of aarnphm/whispercpp)
  • aarnphm/whispercpp (Pybind11)
  • abdeladim-s/pywhispercpp (Pybind11)
• R: bnosac/audio.whisper
• Unity: macoron/whisper.unity

The XCFramework is a precompiled version of the library for iOS, visionOS, tvOS, and macOS. It can be used in Swift projects without the need to compile the library from source. For example, the v1.7.5 version of the XCFramework can be used as follows:

// swift-tools-version: 5.10
// The swift-tools-version declares the minimum version of Swift required to build this package.

import PackageDescription

let package = Package(
    name: "Whisper",
    targets: [
        .executableTarget(
            name: "Whisper",
            dependencies: [
                "WhisperFramework"
            ]),
        .binaryTarget(
            name: "WhisperFramework",
            url: "https://github.com/ggml-org/whisper.cpp/releases/download/v1.7.5/whisper-v1.7.5-xcframework.zip",
            checksum: "c7faeb328620d6012e130f3d705c51a6ea6c995605f2df50f6e1ad68c59c6c4a"
        )
    ]
)

Support for Voice Activity Detection (VAD) can be enabled using the --vad argument to whisper-cli. In addition to this option a VAD model is also required.

The way this works is that first the audio samples are passed through the VAD model which will detect speech segments. Using this information the only the speech segments that are detected are extracted from the original audio input and passed to whisper for processing. This reduces the amount of audio data that needs to be processed by whisper and can significantly speed up the transcription process.

The following VAD models are currently supported:

Silero-vad is a lightweight VAD model written in Python that is fast and accurate.

Models can be downloaded by running the following command on Linux or MacOS:

$ ./models/download-vad-model.sh silero-v5.1.2
Downloading ggml model silero-v5.1.2 from 'https://huggingface.co/ggml-org/whisper-vad' ...
ggml-silero-v5.1.2.bin        100%[==============================================>] 864.35K  --.-KB/s    in 0.04s
Done! Model 'silero-v5.1.2' saved in '/path/models/ggml-silero-v5.1.2.bin'
You can now use it like this:

  $ ./build/bin/whisper-cli -vm /path/models/ggml-silero-v5.1.2.bin --vad -f samples/jfk.wav -m models/ggml-base.en.bin

And the following command on Windows:

> .\models\download-vad-model.cmd silero-v5.1.2
Downloading vad model silero-v5.1.2...
Done! Model silero-v5.1.2 saved in C:\Users\danie\work\ai\whisper.cpp\ggml-silero-v5.1.2.bin
You can now use it like this:

C:\path\build\bin\Release\whisper-cli.exe -vm C:\path\ggml-silero-v5.1.2.bin --vad -m models/ggml-base.en.bin -f samples\jfk.wav

To see a list of all available models, run the above commands without any arguments.

This model can be also be converted manually to ggml using the following command:

$ python3 -m venv venv && source venv/bin/activate
$ (venv) pip install silero-vad
$ (venv) $ python models/convert-silero-vad-to-ggml.py --output models/silero.bin
Saving GGML Silero-VAD model to models/silero-v5.1.2-ggml.bin

And it can then be used with whisper as follows:

$ ./build/bin/whisper-cli \
   --file ./samples/jfk.wav \
   --model ./models/ggml-base.en.bin \
   --vad \
   --vad-model ./models/silero-v5.1.2-ggml.bin

• --vad-threshold: Threshold probability for speech detection. A probability for a speech segment/frame above this threshold will be considered as speech.

• --vad-min-speech-duration-ms: Minimum speech duration in milliseconds. Speech segments shorter than this value will be discarded to filter out brief noise or false positives.

• --vad-min-silence-duration-ms: Minimum silence duration in milliseconds. Silence periods must be at least this long to end a speech segment. Shorter silence periods will be ignored and included as part of the speech.

• --vad-max-speech-duration-s: Maximum speech duration in seconds. Speech segments longer than this will be automatically split into multiple segments at silence points exceeding 98ms to prevent excessively long segments.

• --vad-speech-pad-ms: Speech padding in milliseconds. Adds this amount of padding before and after each detected speech segment to avoid cutting off speech edges.

• --vad-samples-overlap: Amount of audio to extend from each speech segment into the next one, in seconds (e.g., 0.10 = 100ms overlap). This ensures speech isn't cut off abruptly between segments when they're concatenated together.

There are various examples of using the library for different projects in the examples folder. Some of the examples are even ported to run in the browser using WebAssembly. Check them out!

If you have any kind of feedback about this project feel free to use the Discussions section and open a new topic. You can use the Show and tell category to share your own projects that use whisper.cpp. If you have a question, make sure to check the Frequently asked questions (#126) discussion.

• Pluggable LLM backend

  • A tiny interface (llm_backend.h) decouples dialogue logic from the actual LLM.
  • backend_lmstudio.{h,cpp} implements this interface for LM Studio (OpenAI-compatible HTTP API).

• Talk to any local model via LM Studio

  • Instead of running a local llama.cpp model, responses can be produced by any model you’ve loaded in LM Studio (e.g. openai/gpt-oss-20b, Llama 3.x, Mistral, Qwen, …).

• New CLI flags for language, streaming, token limits (ASR vs LLM), sanitizing (umlauts), voice id, and sampling.

• DE/EN prompting & language control

  • Whisper ASR language via -l, LLM reply language via --llm-lang (e.g. de).

• Robust output cleaning

  • Removes stray Assistant: / LLaMA: / User: labels to avoid double speaker tags.

• Separate token limits

  • --max-tokens (ASR chunk size, Whisper) vs --llm-max-tokens (LLM reply length).

• Optional streaming

  • With --llm-stream, LLM tokens stream into the console; TTS still plays the final full answer (no half sentences).

examples/talk-llama/
 ├─ talk-llama.cpp          # Main app (audio/VAD, ASR, LLM call, TTS, dialogue loop)
 ├─ llm_backend.h           # Tiny interface for pluggable LLM backends
 ├─ backend_lmstudio.h/.cpp # LM Studio implementation (OpenAI /v1/chat/completions)
 └─ (optionally other backends in the future)

struct LLMToken {
    std::string text;
    bool is_final = false;
};

struct LLMGenerateParams {
    int   max_tokens   = 256;
    float temperature  = 0.7f;
    int   top_k        = 40;
    float top_p        = 0.95f;
    float min_p        = 0.05f;
    int   seed         = -1;
    bool  stream       = true;
    std::string system_prompt;
};

class LLMBackend {
public:
    virtual ~LLMBackend() = default;
    virtual bool init() = 0;
    virtual void shutdown() = 0;
    virtual bool generate(const std::string& prompt,
                          const LLMGenerateParams& params,
                          const std::function<void(const LLMToken&)>& on_token) = 0;
};

• Talks to /v1/chat/completions (OpenAI-compatible).
• Supports non-stream (single JSON) and SSE streaming (incremental tokens).
• Passes sampling params (temperature, top_k, top_p, min_p, seed), system_prompt, and optional stop sequences.
• Designed to be robust to early/empty stops and to strip accidental speaker labels in the reply.

LM Studio (LLM)

• --llm-backend lmstudio – enable LM Studio backend.
• --lmstudio-url URL – e.g. http://localhost:1234/v1.
• --lmstudio-model ID – exact model id as reported by LM Studio (/v1/models).
• --lmstudio-api-key KEY – default lm-studio.
• --lmstudio-test "PROMPT" – single chat call to LM Studio, then exit (smoke test).
• --llm-lang CODE – reply language (de, en); controls the system instruction.
• --llm-max-tokens N – LLM reply token limit (separate from ASR).
• --llm-stream – stream LLM tokens to console (TTS still uses the final text).

ASR / general

• -l LANG, --language LANG – Whisper language (de, en, auto, …).
• -mw FILE – Whisper model file (e.g. models/ggml-large-v3-turbo.bin).
• --allow-umlauts – relaxed sanitizing that keeps äöüÄÖÜß.
• --voice-id N – forwarded to your speak (TTS) script/engine.

Sampling (also applied to LM Studio)

• --temp <f>
• --top-k <n>
• --top-p <f>
• --min-p <f>
• --seed <n>

Important:
--max-tokens = Whisper ASR chunk length (input side)
--llm-max-tokens = LLM answer length (output side)

You can still set persona and bot labels:

• -p, --person NAME (e.g. -p Jens)
• -bn, --bot-name NAME (display only; LM Studio replies are label-free by design)

• Two starter dialogue prompts (EN/DE) shape style and brevity.
• For LM Studio we also push a system instruction:
  • If --llm-lang de: “Always reply in German. Do not include speaker prefixes (like ‘User:’ or ‘Assistant:’). Answer with plain content.”
  • If --llm-lang en: equivalent instruction in English.
• LM Studio dialogue context is intentionally minimal (user text + answer) to reduce label hallucinations and keep the model focused.

• Strips leading Assistant: / <BotName>: / <Person>: if the model prepends them anyway.
• Removes in-line occurrences like \nAssistant: / \nLLaMA: / \n<name>: that occasionally appear in longer replies.
• Removes bracketed meta ([...], (...)) from the ASR input before sending to the LLM.
• --allow-umlauts keeps äöüÄÖÜß in ASR text; without it, sanitizing is stricter.

• With --llm-stream, tokens are printed as they arrive for a snappier feel.
• TTS still plays the completed answer (no mid-sentence playback), to keep audio natural.

• If you do not enable LM Studio, the classic llama.cpp path runs:
  • Prompt eval, sampler chain (top-k/top-p/temp/min-p/seed), session caching, anti-prompts, etc.

Extra deps for LM Studio backend:

• libcurl (HTTP + SSE)
• nlohmann/json (header-only)

macOS (Homebrew):

brew install curl nlohmann-json

find_package(CURL REQUIRED)
# nlohmann/json is header-only

target_sources(whisper-talk-llama PRIVATE
    ${CMAKE_CURRENT_SOURCE_DIR}/backend_lmstudio.cpp
    ${CMAKE_CURRENT_SOURCE_DIR}/llm_backend.h
    ${CMAKE_CURRENT_SOURCE_DIR}/backend_lmstudio.h
)

target_link_libraries(whisper-talk-llama PRIVATE CURL::libcurl)

Build:

cmake -S . -B build -DWHISPER_ALL_WARNINGS=OFF
cmake --build build -j

Start server:

lms server start
# => "Server is now running on port 1234"

List models:

curl -s http://localhost:1234/v1/models | jq -r '.data[].id'

./build/bin/whisper-talk-llama \
  --llm-backend lmstudio \
  --lmstudio-url http://localhost:1234/v1 \
  --lmstudio-model "openai/gpt-oss-20b" \
  -mw models/ggml-large-v3-turbo.bin \
  -l de \
  --llm-lang de \
  --allow-umlauts \
  --voice-id 2 \
  --top-p 0.95 --temp 0.3 \
  --max-tokens 128 \
  --llm-max-tokens 256 \
  --llm-stream

If answers cut off too early:

• Increase --llm-max-tokens (e.g. 512–1024).
• Lower --temp (e.g. 0.2–0.3) for crisper, shorter text.

To change the user name in transcripts:

-p Jens

• Empty reply due to early stop
  Some models may stop at position 0 when a stop sequence matches. The backend retries once without stop. If repeated, run without stop sequences.

• Doubled labels like LLaMA: / Jens:
  The cleaner strips leading and inline labels; also ensure you set --llm-lang so the system instruction tells the model to omit prefixes.

• Weird characters / lost umlauts
  Use --allow-umlauts to keep äöüÄÖÜß during sanitizing.

• Build errors (curl/json)
  Ensure libcurl is installed and linked; ensure nlohmann/json headers are available.

• ASR vs LLM token limits confusion
  --max-tokens is the ASR chunk size (input).
  --llm-max-tokens is the LLM output length (answer).

LM Studio

• --llm-backend lmstudio
• --lmstudio-url <URL> (e.g. http://localhost:1234/v1)
• --lmstudio-model <ID> (from /v1/models)
• --lmstudio-api-key <KEY> (default lm-studio)
• --lmstudio-test "<PROMPT>"
• --llm-lang {de|en}
• --llm-max-tokens N
• --llm-stream

Sampling

• --temp <f>
• --top-k <n>
• --top-p <f>
• --min-p <f>
• --seed <n>

ASR / general

• -mw <file> (Whisper model)
• -l <lang> (Whisper: de, en, auto, …)
• --max-tokens <n> (ASR chunk)
• -vms <ms> (voice window)
• -ac <n> (audio ctx)
• --allow-umlauts
• --voice-id <n>
• -p <name> (person) / -bn <name> (bot name)
• -t <n> / -ngl <n> / --flash-attn / -ng

• Pluggable backend: easy to swap LM Studio, ollama, or remote APIs.
• Speech-first UX: separate control of ASR input language and LLM reply language.
• Stable prompting: system instructions + cleaning reduce label echo.
• More control: distinct token limits, streaming, sanitizing, voice-id for TTS.