Split single Voxtral transcript segments

internal/transcription/client.go

@@ -377,10 +377,7 @@ func (c *Client) transcribeWithProvider(ctx context.Context, provider ProviderCo
 			return nil, attempt, err
 		}
 		text := strings.TrimSpace(resp.Text)
-		segments := resp.Segments
-		if len(segments) == 0 {
-			segments = segmentTranscriptText(text, in.Diarize)
-		}
+		segments := normalizeAudioLLMSegments(resp.Segments, text, in.Diarize)
 		attempt.Status = "ok"
 		attempt.Model = resp.Model
 		attempt.Text = text
@@ -749,6 +746,42 @@ func convertAudioSegments(in []audioTranscriptionSegment) []Segment {
 	return out
 }
 
+func normalizeAudioLLMSegments(segments []Segment, text string, diarize bool) []Segment {
+	text = strings.TrimSpace(text)
+	if len(segments) <= 1 && text != "" {
+		heuristic := segmentTranscriptText(text, diarize)
+		if len(heuristic) > len(segments) {
+			segments = heuristic
+		}
+	}
+	return ensureHeuristicSpeakers(segments, diarize)
+}
+
+func ensureHeuristicSpeakers(segments []Segment, diarize bool) []Segment {
+	if !diarize || len(segments) < 2 || segmentsHaveSpeakers(segments) {
+		return segments
+	}
+	out := make([]Segment, len(segments))
+	copy(out, segments)
+	for i := range out {
+		if i%2 == 0 {
+			out[i].Speaker = "SPEAKER_00"
+		} else {
+			out[i].Speaker = "SPEAKER_01"
+		}
+	}
+	return out
+}
+
+func segmentsHaveSpeakers(segments []Segment) bool {
+	for _, segment := range segments {
+		if strings.TrimSpace(segment.Speaker) != "" {
+			return true
+		}
+	}
+	return false
+}
+
 func segmentTranscriptText(text string, diarize bool) []Segment {
 	parts := splitTranscriptSentences(text)
 	out := make([]Segment, 0, len(parts))

internal/transcription/client_test.go

@@ -135,6 +135,30 @@ func TestSegmentTranscriptTextAddsHeuristicSpeakers(t *testing.T) {
 	}
 }
 
+func TestNormalizeAudioLLMSegmentsSplitsSingleLongSegment(t *testing.T) {
+	text := "Алло, добрый день. Да, слушаю. Скажите, квартира продается? Да, продается."
+	got := normalizeAudioLLMSegments([]Segment{{Start: 0, End: 12, Text: text}}, text, true)
+	if len(got) < 2 {
+		t.Fatalf("segments = %#v, want heuristic split", got)
+	}
+	if got[0].Speaker != "SPEAKER_00" || got[1].Speaker != "SPEAKER_01" {
+		t.Fatalf("speakers = %q/%q", got[0].Speaker, got[1].Speaker)
+	}
+}
+
+func TestNormalizeAudioLLMSegmentsKeepsSegmentsAndAddsSpeakers(t *testing.T) {
+	got := normalizeAudioLLMSegments([]Segment{
+		{Start: 0, End: 1, Text: "Алло."},
+		{Start: 1, End: 2, Text: "Да, слушаю."},
+	}, "Алло. Да, слушаю.", true)
+	if len(got) != 2 {
+		t.Fatalf("segments = %#v", got)
+	}
+	if got[0].Speaker != "SPEAKER_00" || got[1].Speaker != "SPEAKER_01" {
+		t.Fatalf("speakers = %q/%q", got[0].Speaker, got[1].Speaker)
+	}
+}
+
 func near(got, want float64) bool {
 	return math.Abs(got-want) < 0.000001
 }