Browse Source

Ship deterministic speech fixture for VAD regressions

master
evgeny 2 days ago
parent
commit
c9131fff39
  1. 4
      tests/Makefile.am
  2. BIN
      tests/data/silero_speech.pcm
  3. 12
      tests/data/silero_speech.txt

4
tests/Makefile.am

@ -1,5 +1,7 @@
# Tests Makefile.am for utun - all tests with new ll_queue library
EXTRA_DIST = data/silero_speech.pcm data/silero_speech.txt
# All available tests (check_PROGRAMS runs via automake check-TESTS)
check_PROGRAMS = \
test_ll_queue \
@ -560,7 +562,7 @@ test_radio_vad_SOURCES = test_radio_vad.c
test_radio_vad_LDADD = -lm
test_silero_vad_SOURCES = test_silero_vad.c
test_silero_vad_CFLAGS = -I$(top_srcdir)/lib
test_silero_vad_CFLAGS = -I$(top_srcdir)/lib -DTEST_SILERO_SPEECH_PATH='"$(srcdir)/data/silero_speech.pcm"'
test_silero_vad_LDADD = $(COMMON_LIBS) @SILERO_VAD_LIBS@
test_speex_aec_SOURCES = test_speex_aec.c

BIN
tests/data/silero_speech.pcm

Binary file not shown.

12
tests/data/silero_speech.txt

@ -0,0 +1,12 @@
silero_speech.pcm — синтезированная речь для регрессии Silero VAD.
Формат: mono, 16000 Hz, signed int16 little endian; 16384 отсчёта (32 окна по 512).
Фраза: "This is a speech recognition test. The radio should start transmitting when I speak. One two three four five."
Источник: встроенный синтезатор flite в FFmpeg, голос по умолчанию (kal).
Файл хранится в репозитории; FFmpeg/flite при запуске теста не требуются.
Воспроизведение файла:
ffmpeg -f s16le -ar 16000 -ac 1 -i silero_speech.pcm silero_speech.wav
Генерация исходного сигнала:
ffmpeg -f lavfi -i "flite=text='This is a speech recognition test. The radio should start transmitting when I speak. One two three four five.'" -ar 16000 -ac 1 -f s16le speech_full.pcm
silero_speech.pcm содержит первые 32768 байт speech_full.pcm.
Loading…
Cancel
Save