Feat:new api /sequence2txt and update QWenSeq2txt (#11643)

### What problem does this PR solve?
change:
new api /sequence2txt,
update QWenSeq2txt and ZhipuSeq2txt

### Type of change

- [x] New Feature (non-breaking change which adds functionality)

api/apps/conversation_app.py

@@ -14,9 +14,11 @@
 #  limitations under the License.
 #
 import json
+import os
 import re
 import logging
 from copy import deepcopy
+import tempfile
 from quart import Response, request
 from api.apps import current_user, login_required
 from api.db.db_models import APIToken
@@ -248,6 +250,64 @@ async def completion():
     except Exception as e:
         return server_error_response(e)
 
+@manager.route("/sequence2txt", methods=["POST"])  # noqa: F821
+@login_required
+async def sequence2txt():
+    req = await request.form
+    stream_mode = req.get("stream", "false").lower() == "true"
+    files = await request.files
+    if "file" not in files:
+        return get_data_error_result(message="Missing 'file' in multipart form-data")
+
+    uploaded = files["file"]
+
+    ALLOWED_EXTS = {
+        ".wav", ".mp3", ".m4a", ".aac",
+        ".flac", ".ogg", ".webm",
+        ".opus", ".wma"
+    }
+
+    filename = uploaded.filename or ""
+    suffix = os.path.splitext(filename)[-1].lower()
+    if suffix not in ALLOWED_EXTS:
+        return get_data_error_result(message=
+            f"Unsupported audio format: {suffix}. "
+            f"Allowed: {', '.join(sorted(ALLOWED_EXTS))}"
+        )
+    fd, temp_audio_path = tempfile.mkstemp(suffix=suffix)
+    os.close(fd)
+    await uploaded.save(temp_audio_path)
+
+    tenants = TenantService.get_info_by(current_user.id)
+    if not tenants:
+        return get_data_error_result(message="Tenant not found!")
+
+    asr_id = tenants[0]["asr_id"]
+    if not asr_id:
+        return get_data_error_result(message="No default ASR model is set")
+
+    asr_mdl=LLMBundle(tenants[0]["tenant_id"], LLMType.SPEECH2TEXT, asr_id)
+    if not stream_mode:
+        text = asr_mdl.transcription(temp_audio_path)
+        try:
+            os.remove(temp_audio_path)
+        except Exception as e:
+            logging.error(f"Failed to remove temp audio file: {str(e)}")
+        return get_json_result(data={"text": text})
+    async def event_stream():
+        try:
+            for evt in asr_mdl.stream_transcription(temp_audio_path):
+                yield f"data: {json.dumps(evt, ensure_ascii=False)}\n\n"
+        except Exception as e:
+            err = {"event": "error", "text": str(e)}
+            yield f"data: {json.dumps(err, ensure_ascii=False)}\n\n"
+        finally:
+            try:
+                os.remove(temp_audio_path)
+            except Exception as e:
+                logging.error(f"Failed to remove temp audio file: {str(e)}")
+
+    return Response(event_stream(), content_type="text/event-stream")
 
 @manager.route("/tts", methods=["POST"])  # noqa: F821
 @login_required

api/db/services/llm_service.py

@@ -185,6 +185,66 @@ class LLMBundle(LLM4Tenant):
 
         return txt
 
+    def stream_transcription(self, audio):
+        mdl = self.mdl
+        supports_stream = hasattr(mdl, "stream_transcription") and callable(getattr(mdl, "stream_transcription"))
+        if supports_stream:
+            if self.langfuse:
+                generation = self.langfuse.start_generation(
+                    trace_context=self.trace_context,
+                    name="stream_transcription",
+                    metadata={"model": self.llm_name}
+                )
+            final_text = ""
+            used_tokens = 0
+
+            try:
+                for evt in mdl.stream_transcription(audio):
+                    if evt.get("event") == "final":
+                        final_text = evt.get("text", "")
+
+                    yield evt
+
+            except Exception as e:
+                err = {"event": "error", "text": str(e)}
+                yield err
+                final_text = final_text or ""
+            finally:
+                if final_text:
+                    used_tokens = num_tokens_from_string(final_text)
+                    TenantLLMService.increase_usage(self.tenant_id, self.llm_type, used_tokens)
+
+                if self.langfuse:
+                    generation.update(
+                        output={"output": final_text},
+                        usage_details={"total_tokens": used_tokens}
+                    )
+                    generation.end()
+
+            return
+
+        if self.langfuse:
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="stream_transcription", metadata={"model": self.llm_name})
+        full_text, used_tokens = mdl.transcription(audio)
+        if not TenantLLMService.increase_usage(
+            self.tenant_id, self.llm_type, used_tokens
+        ):
+            logging.error(
+                f"LLMBundle.stream_transcription can't update token usage for {self.tenant_id}/SEQUENCE2TXT used_tokens: {used_tokens}"
+            )
+        if self.langfuse:
+            generation.update(
+                output={"output": full_text},
+                usage_details={"total_tokens": used_tokens}
+            )
+            generation.end()
+
+        yield {
+            "event": "final",
+            "text": full_text,
+            "streaming": False
+        }
+
     def tts(self, text: str) -> Generator[bytes, None, None]:
         if self.langfuse:
             generation = self.langfuse.start_generation(trace_context=self.trace_context, name="tts", input={"text": text})