admin/ai-service
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Use Voxtral JSON transcription without fake speakers
Some checks failed
CI / test (push) Failing after 9s
Details
Build and Deploy / build-and-deploy (push) Successful in 23s
Details

Browse Source
This commit is contained in:
Grendgi
2026-06-09 22:09:27 +03:00
parent e074f6b226
commit bf945e05e3
2 changed files with 68 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
internal/transcription/client_test.go
View File

@@ -66,21 +66,21 @@ func TestVoxtralUsesAudioTranscriptionsEndpoint(t *testing.T) {
if gotModel != "mistralai/Voxtral-Small-24B-2507" {
t.Fatalf("model = %q", gotModel)
}
if gotResponseFormat != "verbose_json" {
t.Fatalf("response_format = %q, want verbose_json", gotResponseFormat)
if gotResponseFormat != "json" {
t.Fatalf("response_format = %q, want json", gotResponseFormat)
}
if len(got.Segments) != 2 || got.Segments[0].Text != "Алло, тест." || got.Segments[1].Start != 1.2 {
t.Fatalf("segments = %#v", got.Segments)
}
}
func TestSegmentTranscriptTextAddsHeuristicSpeakers(t *testing.T) {
got := segmentTranscriptText("Алло, добрый день. Да, слушаю. Скажите, квартира продается? Да, продается.", true)
func TestSegmentTranscriptTextDoesNotInventSpeakers(t *testing.T) {
got := segmentTranscriptText("Алло, добрый день. Да, слушаю. Скажите, квартира продается? Да, продается.")
if len(got) < 2 {
t.Fatalf("segments = %#v, want multiple", got)
}
if got[0].Speaker != "SPEAKER_00" || got[1].Speaker != "SPEAKER_01" {
t.Fatalf("speakers = %q/%q", got[0].Speaker, got[1].Speaker)
if got[0].Speaker != "" || got[1].Speaker != "" {
t.Fatalf("speakers = %q/%q, want empty", got[0].Speaker, got[1].Speaker)
}
if got[1].Start <= got[0].Start {
t.Fatalf("segment times did not advance: %#v", got)
@@ -93,12 +93,12 @@ func TestNormalizeAudioLLMSegmentsSplitsSingleLongSegment(t *testing.T) {
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)
if got[0].Speaker != "" || got[1].Speaker != "" {
t.Fatalf("speakers = %q/%q, want empty", got[0].Speaker, got[1].Speaker)
}
}
func TestNormalizeAudioLLMSegmentsKeepsSegmentsAndAddsSpeakers(t *testing.T) {
func TestNormalizeAudioLLMSegmentsKeepsSegmentsWithoutFakeSpeakers(t *testing.T) {
got := normalizeAudioLLMSegments([]Segment{
{Start: 0, End: 1, Text: "Алло."},
{Start: 1, End: 2, Text: "Да, слушаю."},
@@ -106,7 +106,21 @@ func TestNormalizeAudioLLMSegmentsKeepsSegmentsAndAddsSpeakers(t *testing.T) {
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)
if got[0].Speaker != "" || got[1].Speaker != "" {
t.Fatalf("speakers = %q/%q, want empty", got[0].Speaker, got[1].Speaker)
}
}
func TestNormalizeAudioLLMSegmentsUsesExplicitSpeakerLabels(t *testing.T) {
text := "Спикер 1: Алло, добрый день. Спикер 2: Да, слушаю. Спикер 1: Скажите, квартира продается?"
got := normalizeAudioLLMSegments(nil, text, true)
if len(got) != 3 {
t.Fatalf("segments = %#v, want 3", got)
}
if got[0].Speaker != "SPEAKER_00" || got[1].Speaker != "SPEAKER_01" || got[2].Speaker != "SPEAKER_00" {
t.Fatalf("speakers = %q/%q/%q", got[0].Speaker, got[1].Speaker, got[2].Speaker)
}
if got[0].Text != "Алло, добрый день." || got[1].Text != "Да, слушаю." {
t.Fatalf("texts = %#v", got)
}
}