diff --git a/frontend/src/App.jsx b/frontend/src/App.jsx
index 9856dc6..fab5414 100644
--- a/frontend/src/App.jsx
+++ b/frontend/src/App.jsx
@@ -2,7 +2,8 @@ import React, { useEffect, useMemo, useRef, useState } from 'react'
 
 const API_BASE = '/api'
 const AUTO_STOP_SILENCE_MS = 2500
-const SPEECH_START_THRESHOLD = 0.05
+const SPEECH_START_THRESHOLD = 0.02
+const SILENCE_THRESHOLD = 0.012
 
 async function parseApiResponse(res) {
   const contentType = res.headers.get('content-type') || ''
@@ -90,6 +91,7 @@ export default function App() {
   const [isTranscribing, setIsTranscribing] = useState(false)
   const [voiceStatus, setVoiceStatus] = useState('')
   const [errorMessage, setErrorMessage] = useState('')
+  const [micLevel, setMicLevel] = useState(0)
   const [voices, setVoices] = useState([])
   const [selectedVoiceURI, setSelectedVoiceURI] = useState('')
   const mediaRecorderRef = useRef(null)
@@ -101,6 +103,7 @@ export default function App() {
   const sourceNodeRef = useRef(null)
   const animationFrameRef = useRef(null)
   const silenceStartedAtRef = useRef(null)
+  const hasSpeechInSegmentRef = useRef(false)
   const isRecordingRef = useRef(false)
   const isAutoListeningRef = useRef(false)
   const isTranscribingRef = useRef(false)
@@ -285,6 +288,8 @@ export default function App() {
     }
     analyserRef.current = null
     silenceStartedAtRef.current = null
+    hasSpeechInSegmentRef.current = false
+    setMicLevel(0)
     if (mediaStreamRef.current) {
       mediaStreamRef.current.getTracks().forEach((track) => track.stop())
       mediaStreamRef.current = null
@@ -370,6 +375,7 @@ export default function App() {
     recordedChunksRef.current = []
     recordingMimeTypeRef.current = mimeType || recorder.mimeType || 'audio/webm'
     silenceStartedAtRef.current = null
+    hasSpeechInSegmentRef.current = false
 
     recorder.ondataavailable = (event) => {
       if (event.data && event.data.size > 0) {
@@ -384,11 +390,13 @@ export default function App() {
       mediaRecorderRef.current = null
       setIsRecording(false)
 
-      if (audioBlob.size > 0) {
+      if (audioBlob.size > 0 && hasSpeechInSegmentRef.current) {
         await transcribeRecording(audioBlob, finalMimeType, { autoSend: true })
       } else if (isAutoListeningRef.current) {
         setVoiceStatus('Micro actif. Parle quand tu veux.')
       }
+
+      hasSpeechInSegmentRef.current = false
     }
 
     recorder.onerror = () => {
@@ -415,18 +423,19 @@ export default function App() {
 
     const volume = calculateVolume(analyserRef.current)
     const now = Date.now()
+    setMicLevel(volume)
 
     if (volume >= SPEECH_START_THRESHOLD) {
+      hasSpeechInSegmentRef.current = true
       silenceStartedAtRef.current = null
-      if (!isRecordingRef.current && !isTranscribingRef.current) {
-        startSegmentRecording()
-      }
-    } else if (isRecordingRef.current) {
+    } else if (isRecordingRef.current && hasSpeechInSegmentRef.current && volume <= SILENCE_THRESHOLD) {
       if (!silenceStartedAtRef.current) {
         silenceStartedAtRef.current = now
       } else if (now - silenceStartedAtRef.current >= AUTO_STOP_SILENCE_MS) {
         stopSegmentRecording()
       }
+    } else if (volume > SILENCE_THRESHOLD) {
+      silenceStartedAtRef.current = null
     }
 
     animationFrameRef.current = requestAnimationFrame(monitorMicrophone)
@@ -459,6 +468,7 @@ export default function App() {
       mediaStreamRef.current = stream
       setIsAutoListening(true)
       setVoiceStatus('Micro actif. Parle quand tu veux, j’enverrai après 2,5 s de silence.')
+      await startSegmentRecording()
       monitorMicrophone()
     } catch {
       setVoiceStatus('Accès au micro refusé ou indisponible.')
@@ -603,6 +613,11 @@ export default function App() {
           <button type="submit">Envoyer</button>
         </form>
 
+        {isAutoListening && (
+          <div className="voice-status">
+            Niveau micro: {Math.round(Math.min(micLevel * 1200, 100))}%
+          </div>
+        )}
         {voiceStatus && <p className="muted voice-status">{voiceStatus}</p>}
         {errorMessage && <p className="muted voice-status">{errorMessage}</p>}
       </main>