feat(qwen-asr): add support to qwen-asr (#8281)

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2026-01-31 17:52:51 -05:00 · 2026-01-29 21:50:35 +01:00
parent dd8e74a486
commit 1e08e02598
22 changed files with 651 additions and 2 deletions
--- a/.github/workflows/backend.yml
+++ b/.github/workflows/backend.yml
@@ -105,6 +105,19 @@ jobs:
            dockerfile: "./backend/Dockerfile.python"
            context: "./"
            ubuntu-version: '2404'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "9"
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-nvidia-cuda-12-qwen-asr'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "qwen-asr"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
          - build-type: 'cublas'
            cuda-major-version: "12"
            cuda-minor-version: "9"
@@ -366,6 +379,19 @@ jobs:
            dockerfile: "./backend/Dockerfile.python"
            context: "./"
            ubuntu-version: '2404'
+          - build-type: 'cublas'
+            cuda-major-version: "13"
+            cuda-minor-version: "0"
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-nvidia-cuda-13-qwen-asr'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "qwen-asr"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
          - build-type: 'cublas'
            cuda-major-version: "13"
            cuda-minor-version: "0"
@@ -470,6 +496,19 @@ jobs:
            backend: "vibevoice"
            dockerfile: "./backend/Dockerfile.python"
            context: "./"
+          - build-type: 'l4t'
+            cuda-major-version: "13"
+            cuda-minor-version: "0"
+            platforms: 'linux/arm64'
+            tag-latest: 'auto'
+            tag-suffix: '-nvidia-l4t-cuda-13-arm64-qwen-asr'
+            runs-on: 'ubuntu-24.04-arm'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            ubuntu-version: '2404'
+            backend: "qwen-asr"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
          - build-type: 'l4t'
            cuda-major-version: "13"
            cuda-minor-version: "0"
@@ -732,6 +771,19 @@ jobs:
            dockerfile: "./backend/Dockerfile.python"
            context: "./"
            ubuntu-version: '2404'
+          - build-type: 'hipblas'
+            cuda-major-version: ""
+            cuda-minor-version: ""
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-rocm-hipblas-qwen-asr'
+            runs-on: 'arc-runner-set'
+            base-image: "rocm/dev-ubuntu-24.04:6.4.4"
+            skip-drivers: 'false'
+            backend: "qwen-asr"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
          - build-type: 'hipblas'
            cuda-major-version: ""
            cuda-minor-version: ""
@@ -889,6 +941,19 @@ jobs:
            dockerfile: "./backend/Dockerfile.python"
            context: "./"
            ubuntu-version: '2204'
+          - build-type: 'l4t'
+            cuda-major-version: "12"
+            cuda-minor-version: "0"
+            platforms: 'linux/arm64'
+            tag-latest: 'auto'
+            tag-suffix: '-nvidia-l4t-qwen-asr'
+            runs-on: 'ubuntu-24.04-arm'
+            base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
+            skip-drivers: 'true'
+            backend: "qwen-asr"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2204'
          - build-type: 'l4t'
            cuda-major-version: "12"
            cuda-minor-version: "0"
@@ -968,6 +1033,19 @@ jobs:
            dockerfile: "./backend/Dockerfile.python"
            context: "./"
            ubuntu-version: '2404'
+          - build-type: 'intel'
+            cuda-major-version: ""
+            cuda-minor-version: ""
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-intel-qwen-asr'
+            runs-on: 'arc-runner-set'
+            base-image: "intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04"
+            skip-drivers: 'false'
+            backend: "qwen-asr"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
          - build-type: 'intel'
            cuda-major-version: ""
            cuda-minor-version: ""
@@ -1380,6 +1458,19 @@ jobs:
            dockerfile: "./backend/Dockerfile.python"
            context: "./"
            ubuntu-version: '2404'
+          - build-type: ''
+            cuda-major-version: ""
+            cuda-minor-version: ""
+            platforms: 'linux/amd64,linux/arm64'
+            tag-latest: 'auto'
+            tag-suffix: '-cpu-qwen-asr'
+            runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:24.04"
+            skip-drivers: 'false'
+            backend: "qwen-asr"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./"
+            ubuntu-version: '2404'
          - build-type: ''
            cuda-major-version: ""
            cuda-minor-version: ""
--- a/.github/workflows/test-extra.yml
+++ b/.github/workflows/test-extra.yml
@@ -304,6 +304,25 @@ jobs:
        run: |
          make --jobs=5 --output-sync=target -C backend/python/qwen-tts
          make --jobs=5 --output-sync=target -C backend/python/qwen-tts test
+  tests-qwen-asr:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y build-essential ffmpeg sox
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          # Install UV
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
+      - name: Test qwen-asr
+        run: |
+          make --jobs=5 --output-sync=target -C backend/python/qwen-asr
+          make --jobs=5 --output-sync=target -C backend/python/qwen-asr test
  tests-voxcpm:
    runs-on: ubuntu-latest
    steps:
--- a/8
+++ b/8
@@ -1,5 +1,5 @@
 # Disable parallel execution for backend builds
-.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/piper backends/stablediffusion-ggml backends/whisper backends/faster-whisper backends/silero-vad backends/local-store backends/huggingface backends/rfdetr backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/stablediffusion-ggml-darwin backends/vllm backends/vllm-omni backends/moonshine backends/pocket-tts backends/qwen-tts backends/voxcpm
+.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/piper backends/stablediffusion-ggml backends/whisper backends/faster-whisper backends/silero-vad backends/local-store backends/huggingface backends/rfdetr backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/stablediffusion-ggml-darwin backends/vllm backends/vllm-omni backends/moonshine backends/pocket-tts backends/qwen-tts backends/qwen-asr backends/voxcpm

 GOCMD=go
 GOTEST=$(GOCMD) test
@@ -319,6 +319,7 @@ prepare-test-extra: protogen-python
 	$(MAKE) -C backend/python/moonshine
 	$(MAKE) -C backend/python/pocket-tts
 	$(MAKE) -C backend/python/qwen-tts
+	$(MAKE) -C backend/python/qwen-asr
 	$(MAKE) -C backend/python/voxcpm

 test-extra: prepare-test-extra
@@ -331,6 +332,7 @@ test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/moonshine test
 	$(MAKE) -C backend/python/pocket-tts test
 	$(MAKE) -C backend/python/qwen-tts test
+	$(MAKE) -C backend/python/qwen-asr test
 	$(MAKE) -C backend/python/voxcpm test

 DOCKER_IMAGE?=local-ai
@@ -464,6 +466,7 @@ BACKEND_VIBEVOICE = vibevoice|python|.|--progress=plain|true
 BACKEND_MOONSHINE = moonshine|python|.|false|true
 BACKEND_POCKET_TTS = pocket-tts|python|.|false|true
 BACKEND_QWEN_TTS = qwen-tts|python|.|false|true
+BACKEND_QWEN_ASR = qwen-asr|python|.|false|true
 BACKEND_VOXCPM = voxcpm|python|.|false|true

 # Helper function to build docker image for a backend
@@ -510,13 +513,14 @@ $(eval $(call generate-docker-build-target,$(BACKEND_VIBEVOICE)))
 $(eval $(call generate-docker-build-target,$(BACKEND_MOONSHINE)))
 $(eval $(call generate-docker-build-target,$(BACKEND_POCKET_TTS)))
 $(eval $(call generate-docker-build-target,$(BACKEND_QWEN_TTS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_QWEN_ASR)))
 $(eval $(call generate-docker-build-target,$(BACKEND_VOXCPM)))

 # Pattern rule for docker-save targets
 docker-save-%: backend-images
 	docker save local-ai-backend:$* -o backend-images/$*.tar

-docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-vllm-omni docker-build-transformers docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-chatterbox docker-build-vibevoice docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts docker-build-voxcpm
+docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-vllm-omni docker-build-transformers docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-chatterbox docker-build-vibevoice docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts docker-build-qwen-asr docker-build-voxcpm

 ########################################################
 ### END Backends
--- a/backend/index.yaml
+++ b/backend/index.yaml
@@ -414,6 +414,28 @@
    nvidia-l4t-cuda-12: "nvidia-l4t-qwen-tts"
    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-qwen-tts"
  icon: https://cdn-avatars.huggingface.co/v1/production/uploads/620760a26e3b7210c2ff1943/-s1gyJfvbE1RgO5iBeNOi.png
+- &qwen-asr
+  urls:
+    - https://github.com/QwenLM/Qwen3-ASR
+  description: |
+    Qwen3-ASR is an automatic speech recognition model supporting multiple languages and batch inference.
+  tags:
+    - speech-recognition
+    - ASR
+  license: apache-2.0
+  name: "qwen-asr"
+  alias: "qwen-asr"
+  capabilities:
+    nvidia: "cuda12-qwen-asr"
+    intel: "intel-qwen-asr"
+    amd: "rocm-qwen-asr"
+    nvidia-l4t: "nvidia-l4t-qwen-asr"
+    default: "cpu-qwen-asr"
+    nvidia-cuda-13: "cuda13-qwen-asr"
+    nvidia-cuda-12: "cuda12-qwen-asr"
+    nvidia-l4t-cuda-12: "nvidia-l4t-qwen-asr"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-qwen-asr"
+  icon: https://cdn-avatars.huggingface.co/v1/production/uploads/620760a26e3b7210c2ff1943/-s1gyJfvbE1RgO5iBeNOi.png
 - &voxcpm
  urls:
    - https://github.com/ModelBest/VoxCPM
@@ -1671,6 +1693,89 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-qwen-tts"
  mirrors:
    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-qwen-tts
+## qwen-asr
+- !!merge <<: *qwen-asr
+  name: "qwen-asr-development"
+  capabilities:
+    nvidia: "cuda12-qwen-asr-development"
+    intel: "intel-qwen-asr-development"
+    amd: "rocm-qwen-asr-development"
+    nvidia-l4t: "nvidia-l4t-qwen-asr-development"
+    default: "cpu-qwen-asr-development"
+    nvidia-cuda-13: "cuda13-qwen-asr-development"
+    nvidia-cuda-12: "cuda12-qwen-asr-development"
+    nvidia-l4t-cuda-12: "nvidia-l4t-qwen-asr-development"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-qwen-asr-development"
+- !!merge <<: *qwen-asr
+  name: "cpu-qwen-asr"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-qwen-asr"
+  mirrors:
+    - localai/localai-backends:latest-cpu-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "cpu-qwen-asr-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-qwen-asr"
+  mirrors:
+    - localai/localai-backends:master-cpu-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "cuda12-qwen-asr"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-qwen-asr"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "cuda12-qwen-asr-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-qwen-asr"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "cuda13-qwen-asr"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-qwen-asr"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "cuda13-qwen-asr-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-qwen-asr"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "intel-qwen-asr"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-qwen-asr"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "intel-qwen-asr-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-qwen-asr"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "rocm-qwen-asr"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-qwen-asr"
+  mirrors:
+    - localai/localai-backends:latest-gpu-rocm-hipblas-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "rocm-qwen-asr-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-qwen-asr"
+  mirrors:
+    - localai/localai-backends:master-gpu-rocm-hipblas-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "nvidia-l4t-qwen-asr"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-qwen-asr"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "nvidia-l4t-qwen-asr-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-qwen-asr"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "cuda13-nvidia-l4t-arm64-qwen-asr"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-cuda-13-arm64-qwen-asr"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-cuda-13-arm64-qwen-asr
+- !!merge <<: *qwen-asr
+  name: "cuda13-nvidia-l4t-arm64-qwen-asr-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-qwen-asr"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-qwen-asr
 ## voxcpm
 - !!merge <<: *voxcpm
  name: "voxcpm-development"
--- a/backend/python/qwen-asr/Makefile
+++ b/backend/python/qwen-asr/Makefile
@@ -0,0 +1,25 @@
+.DEFAULT_GOAL := install
+
+.PHONY: qwen-asr
+qwen-asr:
+	bash install.sh
+
+.PHONY: run
+run: qwen-asr
+	@echo "Running qwen-asr..."
+	bash run.sh
+	@echo "qwen-asr run."
+
+.PHONY: test
+test: qwen-asr
+	@echo "Testing qwen-asr..."
+	bash test.sh
+	@echo "qwen-asr tested."
+
+.PHONY: protogen-clean
+protogen-clean:
+	$(RM) backend_pb2_grpc.py backend_pb2.py
+
+.PHONY: clean
+clean: protogen-clean
+	rm -rf venv __pycache__
--- a/backend/python/qwen-asr/backend.py
+++ b/backend/python/qwen-asr/backend.py
@@ -0,0 +1,212 @@
+#!/usr/bin/env python3
+"""
+gRPC server of LocalAI for Qwen3-ASR (transformers backend, non-vLLM).
+"""
+from concurrent import futures
+import time
+import argparse
+import signal
+import sys
+import os
+import backend_pb2
+import backend_pb2_grpc
+import torch
+from qwen_asr import Qwen3ASRModel
+
+import grpc
+
+
+def is_float(s):
+    try:
+        float(s)
+        return True
+    except ValueError:
+        return False
+
+
+def is_int(s):
+    try:
+        int(s)
+        return True
+    except ValueError:
+        return False
+
+
+_ONE_DAY_IN_SECONDS = 60 * 60 * 24
+MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
+
+
+class BackendServicer(backend_pb2_grpc.BackendServicer):
+    def Health(self, request, context):
+        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
+
+    def LoadModel(self, request, context):
+        if torch.cuda.is_available():
+            device = "cuda"
+        else:
+            device = "cpu"
+        mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
+        if mps_available:
+            device = "mps"
+        if not torch.cuda.is_available() and request.CUDA:
+            return backend_pb2.Result(success=False, message="CUDA is not available")
+
+        self.device = device
+        self.options = {}
+
+        for opt in request.Options:
+            if ":" not in opt:
+                continue
+            key, value = opt.split(":", 1)
+            if is_float(value):
+                value = float(value)
+            elif is_int(value):
+                value = int(value)
+            elif value.lower() in ["true", "false"]:
+                value = value.lower() == "true"
+            self.options[key] = value
+
+        model_path = request.Model or "Qwen/Qwen3-ASR-1.7B"
+        default_dtype = torch.bfloat16 if self.device == "cuda" else torch.float32
+        load_dtype = default_dtype
+        if "torch_dtype" in self.options:
+            d = str(self.options["torch_dtype"]).lower()
+            if d == "fp16":
+                load_dtype = torch.float16
+            elif d == "bf16":
+                load_dtype = torch.bfloat16
+            elif d == "fp32":
+                load_dtype = torch.float32
+            del self.options["torch_dtype"]
+
+        self.max_inference_batch_size = self.options.get("max_inference_batch_size", 32)
+        self.max_new_tokens = self.options.get("max_new_tokens", 256)
+
+        forced_aligner = self.options.get("forced_aligner")
+        if forced_aligner is not None and isinstance(forced_aligner, str):
+            forced_aligner = forced_aligner.strip() or None
+        attn_implementation = self.options.get("attn_implementation")
+        if attn_implementation is not None and isinstance(attn_implementation, str):
+            attn_implementation = attn_implementation.strip() or None
+
+        if self.device == "mps":
+            device_map = None
+        elif self.device == "cuda":
+            device_map = "cuda:0"
+        else:
+            device_map = "cpu"
+
+        load_kwargs = dict(
+            dtype=load_dtype,
+            device_map=device_map,
+            max_inference_batch_size=self.max_inference_batch_size,
+            max_new_tokens=self.max_new_tokens,
+        )
+        if attn_implementation:
+            load_kwargs["attn_implementation"] = attn_implementation
+        if forced_aligner:
+            load_kwargs["forced_aligner"] = forced_aligner
+            forced_aligner_kwargs = dict(
+                dtype=load_dtype,
+                device_map=device_map,
+            )
+            if attn_implementation:
+                forced_aligner_kwargs["attn_implementation"] = attn_implementation
+            load_kwargs["forced_aligner_kwargs"] = forced_aligner_kwargs
+
+        try:
+            print(f"Loading Qwen3-ASR from {model_path}", file=sys.stderr)
+            if attn_implementation:
+                print(f"Using attn_implementation: {attn_implementation}", file=sys.stderr)
+            if forced_aligner:
+                print(f"Loading with forced_aligner: {forced_aligner}", file=sys.stderr)
+            self.model = Qwen3ASRModel.from_pretrained(model_path, **load_kwargs)
+            print("Qwen3-ASR model loaded successfully", file=sys.stderr)
+        except Exception as err:
+            print(f"[ERROR] LoadModel failed: {err}", file=sys.stderr)
+            import traceback
+            traceback.print_exc(file=sys.stderr)
+            return backend_pb2.Result(success=False, message=str(err))
+
+        return backend_pb2.Result(message="Model loaded successfully", success=True)
+
+    def AudioTranscription(self, request, context):
+        result_segments = []
+        text = ""
+        try:
+            audio_path = request.dst
+            if not audio_path or not os.path.exists(audio_path):
+                print(f"Error: Audio file not found: {audio_path}", file=sys.stderr)
+                return backend_pb2.TranscriptResult(segments=[], text="")
+
+            language = None
+            if request.language and request.language.strip():
+                language = request.language.strip()
+
+            results = self.model.transcribe(audio=audio_path, language=language)
+
+            if not results:
+                return backend_pb2.TranscriptResult(segments=[], text="")
+
+            r = results[0]
+            text = r.text or ""
+
+            if getattr(r, 'time_stamps', None) and len(r.time_stamps) > 0:
+                for idx, ts in enumerate(r.time_stamps):
+                    start_ms = 0
+                    end_ms = 0
+                    seg_text = text
+                    if isinstance(ts, (list, tuple)) and len(ts) >= 3:
+                        start_ms = int(float(ts[0]) * 1000) if ts[0] is not None else 0
+                        end_ms = int(float(ts[1]) * 1000) if ts[1] is not None else 0
+                        seg_text = ts[2] if len(ts) > 2 and ts[2] is not None else ""
+                    result_segments.append(backend_pb2.TranscriptSegment(
+                        id=idx, start=start_ms, end=end_ms, text=seg_text
+                    ))
+            else:
+                if text:
+                    result_segments.append(backend_pb2.TranscriptSegment(
+                        id=0, start=0, end=0, text=text
+                    ))
+        except Exception as err:
+            print(f"Error in AudioTranscription: {err}", file=sys.stderr)
+            import traceback
+            traceback.print_exc(file=sys.stderr)
+            return backend_pb2.TranscriptResult(segments=[], text="")
+
+        return backend_pb2.TranscriptResult(segments=result_segments, text=text)
+
+
+def serve(address):
+    server = grpc.server(
+        futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),
+        ])
+    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
+    server.add_insecure_port(address)
+    server.start()
+    print("Server started. Listening on: " + address, file=sys.stderr)
+
+    def signal_handler(sig, frame):
+        print("Received termination signal. Shutting down...")
+        server.stop(0)
+        sys.exit(0)
+
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    try:
+        while True:
+            time.sleep(_ONE_DAY_IN_SECONDS)
+    except KeyboardInterrupt:
+        server.stop(0)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Run the gRPC server.")
+    parser.add_argument("--addr", default="localhost:50051", help="The address to bind the server to.")
+    args = parser.parse_args()
+    serve(args.addr)
--- a/backend/python/qwen-asr/install.sh
+++ b/backend/python/qwen-asr/install.sh
@@ -0,0 +1,21 @@
+#!/bin/bash
+set -e
+
+EXTRA_PIP_INSTALL_FLAGS="--no-build-isolation"
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+if [ "x${BUILD_PROFILE}" == "xintel" ]; then
+    EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
+fi
+
+PYTHON_VERSION="3.12"
+PYTHON_PATCH="12"
+PY_STANDALONE_TAG="20251120"
+
+installRequirements
--- a/backend/python/qwen-asr/requirements-cpu.txt
+++ b/backend/python/qwen-asr/requirements-cpu.txt
@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cpu
+torch
+qwen-asr
--- a/backend/python/qwen-asr/requirements-cublas12-after.txt
+++ b/backend/python/qwen-asr/requirements-cublas12-after.txt
@@ -0,0 +1 @@
+flash-attn
--- a/backend/python/qwen-asr/requirements-cublas12.txt
+++ b/backend/python/qwen-asr/requirements-cublas12.txt
@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cu121
+torch
+qwen-asr
--- a/backend/python/qwen-asr/requirements-cublas13.txt
+++ b/backend/python/qwen-asr/requirements-cublas13.txt
@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+torch
+qwen-asr
--- a/backend/python/qwen-asr/requirements-hipblas.txt
+++ b/backend/python/qwen-asr/requirements-hipblas.txt
@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/rocm6.3
+torch==2.7.1+rocm6.3
+qwen-asr
--- a/backend/python/qwen-asr/requirements-intel-after.txt
+++ b/backend/python/qwen-asr/requirements-intel-after.txt
@@ -0,0 +1 @@
+flash-attn
--- a/backend/python/qwen-asr/requirements-intel.txt
+++ b/backend/python/qwen-asr/requirements-intel.txt
@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch
+qwen-asr
--- a/backend/python/qwen-asr/requirements-l4t12.txt
+++ b/backend/python/qwen-asr/requirements-l4t12.txt
@@ -0,0 +1,3 @@
+--extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu129/
+torch
+qwen-asr
--- a/backend/python/qwen-asr/requirements-l4t13.txt
+++ b/backend/python/qwen-asr/requirements-l4t13.txt
@@ -0,0 +1,3 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+torch
+qwen-asr
--- a/backend/python/qwen-asr/requirements-mps.txt
+++ b/backend/python/qwen-asr/requirements-mps.txt
@@ -0,0 +1,2 @@
+torch==2.7.1
+qwen-asr
--- a/backend/python/qwen-asr/requirements.txt
+++ b/backend/python/qwen-asr/requirements.txt
@@ -0,0 +1,5 @@
+grpcio==1.71.0
+protobuf
+certifi
+packaging==24.1
+setuptools
--- a/backend/python/qwen-asr/run.sh
+++ b/backend/python/qwen-asr/run.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+startBackend $@
--- a/backend/python/qwen-asr/test.py
+++ b/backend/python/qwen-asr/test.py
@@ -0,0 +1,94 @@
+"""
+Tests for the Qwen3-ASR gRPC backend.
+"""
+import unittest
+import subprocess
+import time
+import os
+import tempfile
+import shutil
+import backend_pb2
+import backend_pb2_grpc
+
+import grpc
+
+# Skip heavy transcription test in CI (model download + inference)
+SKIP_ASR_TESTS = os.environ.get("SKIP_ASR_TESTS", "false").lower() == "true"
+
+
+class TestBackendServicer(unittest.TestCase):
+    def setUp(self):
+        self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
+        time.sleep(15)
+
+    def tearDown(self):
+        self.service.terminate()
+        self.service.wait()
+
+    def test_server_startup(self):
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.Health(backend_pb2.HealthMessage())
+                self.assertEqual(response.message, b'OK')
+        except Exception as err:
+            print(err)
+            self.fail("Server failed to start")
+        finally:
+            self.tearDown()
+
+    def test_load_model(self):
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.LoadModel(backend_pb2.ModelOptions(Model="Qwen/Qwen3-ASR-1.7B"))
+                self.assertTrue(response.success, response.message)
+                self.assertEqual(response.message, "Model loaded successfully")
+        except Exception as err:
+            print(err)
+            self.fail("LoadModel service failed")
+        finally:
+            self.tearDown()
+
+    @unittest.skipIf(SKIP_ASR_TESTS, "ASR transcription test skipped (SKIP_ASR_TESTS=true)")
+    def test_audio_transcription(self):
+        temp_dir = tempfile.mkdtemp()
+        audio_file = os.path.join(temp_dir, 'audio.wav')
+        try:
+            url = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_en.wav"
+            result = subprocess.run(
+                ["wget", "-q", url, "-O", audio_file],
+                capture_output=True,
+                text=True,
+                timeout=30,
+            )
+            if result.returncode != 0:
+                self.skipTest(f"Could not download sample audio: {result.stderr}")
+            if not os.path.exists(audio_file):
+                self.skipTest("Sample audio file not found after download")
+
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                load_response = stub.LoadModel(backend_pb2.ModelOptions(Model="Qwen/Qwen3-ASR-0.6B"))
+                self.assertTrue(load_response.success, load_response.message)
+
+                transcript_response = stub.AudioTranscription(
+                    backend_pb2.TranscriptRequest(dst=audio_file)
+                )
+                self.assertIsNotNone(transcript_response)
+                self.assertIsNotNone(transcript_response.text)
+                self.assertGreaterEqual(len(transcript_response.segments), 0)
+                all_text = ""
+                for segment in transcript_response.segments:
+                    all_text += segment.text
+                print(f"All text: {all_text}")
+                self.assertIn("big", all_text)
+                if transcript_response.segments:
+                    self.assertIsNotNone(transcript_response.segments[0].text)
+        finally:
+            self.tearDown()
+            if os.path.exists(temp_dir):
+                shutil.rmtree(temp_dir)
--- a/backend/python/qwen-asr/test.sh
+++ b/backend/python/qwen-asr/test.sh
@@ -0,0 +1,11 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+runUnittests
--- a/gallery/index.yaml
+++ b/gallery/index.yaml
@@ -94,6 +94,34 @@
      voice: Aiden # Available speakers: Vivian, Serena, Uncle_Fu, Dylan, Eric, Ryan, Aiden, Ono_Anna, Sohee
    parameters:
      model: Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice
+- &qwen-asr
+  urls:
+    - https://huggingface.co/Qwen/Qwen3-ASR-1.7B
+  description: |
+    Qwen3-ASR is an automatic speech recognition model supporting multiple languages and batch inference.
+  tags:
+    - speech-recognition
+    - ASR
+  license: apache-2.0
+  icon: https://cdn-avatars.huggingface.co/v1/production/uploads/620760a26e3b7210c2ff1943/-s1gyJfvbE1RgO5iBeNOi.png
+  name: "qwen3-asr-1.7b"
+  url: "github:mudler/LocalAI/gallery/virtual.yaml@master"
+  overrides:
+    backend: qwen-asr
+    known_usecases:
+      - transcript
+    parameters:
+      model: Qwen/Qwen3-ASR-1.7B
+- !!merge <<: *qwen-asr
+  urls:
+    - https://huggingface.co/Qwen/Qwen3-ASR-0.6B
+  name: "qwen3-asr-0.6b"
+  overrides:
+    backend: qwen-asr
+    known_usecases:
+      - transcript
+    parameters:
+      model: Qwen/Qwen3-ASR-0.6B
 - name: "huihui-glm-4.7-flash-abliterated-i1"
  url: "github:mudler/LocalAI/gallery/virtual.yaml@master"
  urls: