diff --git a/content/docs/configuration/stt_tts.mdx b/content/docs/configuration/stt_tts.mdx
index a6ea4336d..676922a01 100644
--- a/content/docs/configuration/stt_tts.mdx
+++ b/content/docs/configuration/stt_tts.mdx
@@ -182,15 +182,85 @@ You can specify either the full domain or just the instance name. If you provide
 
 Refer to the OpenAI Whisper section, adjusting the `url` and `model` as needed.
 
-example
-  
 ```yaml
 speech:
   stt:
     openai:
       url: 'http://host.docker.internal:8080/v1/audio/transcriptions'
       model: 'whisper'
-  ```
+```
+
+- #### FunASR with SenseVoice
+
+[FunASR](https://github.com/modelscope/FunASR) provides an OpenAI-compatible transcription endpoint. The following example uses the lightweight SenseVoice model for multilingual speech recognition, including Mandarin, Cantonese, English, Japanese, and Korean.
+
+Create a separate Python environment for the FunASR service:
+
+```bash
+python -m venv .venv
+source .venv/bin/activate
+```
+
+Install matching PyTorch and torchaudio builds for your platform using the [PyTorch installation guide](https://pytorch.org/get-started/locally/). They are required for this CPU example as well as for GPU inference; installing `funasr` alone does not select them for you. Then install the server dependencies in the same environment:
+
+```bash
+pip install "funasr>=1.3.26" fastapi uvicorn python-multipart
+```
+
+Start the server on CPU:
+
+```bash
+funasr-server \
+  --host 0.0.0.0 \
+  --port 8000 \
+  --device cpu \
+  --model sensevoice
+```
+
+Model weights must be cached locally or downloadable when the service loads the model. For NVIDIA GPU inference, select compatible CUDA builds in the PyTorch installation step above, then use `--device cuda`.
+
+<Callout type="warn" title="Restrict access to the transcription server">
+The command above binds all network interfaces so a LibreChat container can reach the host service. FunASR does not require authentication by default. Do not expose this endpoint directly to the public internet: restrict network access and use an authenticated TLS reverse proxy when remote access is required. If both processes run directly on the same host, use `--host 127.0.0.1` instead.
+</Callout>
+
+Verify the endpoint before configuring LibreChat:
+
+```bash
+curl --fail http://localhost:8000/health
+
+curl --fail http://localhost:8000/v1/audio/transcriptions \
+  -F file=@sample.wav \
+  -F model=sensevoice \
+  -F language=zh
+```
+
+A successful request returns JSON containing a `text` field.
+
+Then configure LibreChat:
+
+```yaml
+speech:
+  stt:
+    allowedAddresses:
+      - 'host.docker.internal:8000'
+    openai:
+      url: 'http://host.docker.internal:8000/v1/audio/transcriptions'
+      model: 'sensevoice'
+```
+
+FunASR does not require an API key by default. Add `apiKey` only when an authenticated reverse proxy protects the endpoint.
+
+LibreChat blocks requests to private and loopback addresses unless the exact host and port are exempted under `speech.stt.allowedAddresses`. Use bare `host:port` entries without a scheme or path, and trust only hosts you control. See [Speech SSRF protection](/docs/configuration/librechat_yaml/object_structure/speech#ssrf-protection).
+
+Set both the URL and its matching exemption for your deployment; keep only the entry you need:
+
+Keep the URL scheme `http`, port `8000`, and path `/v1/audio/transcriptions` from the example above. Set the URL host and the single `allowedAddresses` entry as follows:
+
+- LibreChat in Docker and FunASR on the Docker host: URL host `host.docker.internal`, entry `host.docker.internal:8000`
+- Both processes directly on the same host: URL host `127.0.0.1`, entry `127.0.0.1:8000`
+- Both services in the same Compose network: for a service named `funasr`, URL host `funasr`, entry `funasr:8000`; replace the service name and port in both values if they differ
+
+On Linux, add a `host.docker.internal:host-gateway` mapping to the LibreChat container if that hostname is not already available.
 
 
 ## TTS (Text-to-Speech)