From f49cec2ba8830dd8df0ac73a39c118f6e20b06bd Mon Sep 17 00:00:00 2001 From: Paul B Mahol Date: Sun, 5 May 2019 15:01:53 +0200 Subject: [PATCH] avfilter: add asr filter --- Changelog | 1 + configure | 4 + doc/filters.texi | 35 ++++++++ libavfilter/Makefile | 1 + libavfilter/af_asr.c | 181 +++++++++++++++++++++++++++++++++++++++ libavfilter/allfilters.c | 1 + libavfilter/version.h | 2 +- 7 files changed, 224 insertions(+), 1 deletion(-) create mode 100644 libavfilter/af_asr.c diff --git a/Changelog b/Changelog index 1ea7c83324..e6b209ae0a 100644 --- a/Changelog +++ b/Changelog @@ -29,6 +29,7 @@ version : - Support decoding of HEVC 4:4:4 content in vdpau - colorhold filter - xmedian filter +- asr filter version 4.1: diff --git a/configure b/configure index 7130b4c8e9..580af9c0f6 100755 --- a/configure +++ b/configure @@ -307,6 +307,7 @@ External library support: --enable-opengl enable OpenGL rendering [no] --enable-openssl enable openssl, needed for https support if gnutls, libtls or mbedtls is not used [no] + --enable-pocketsphinx enable PocketSphinx, needed for asr filter [no] --disable-sndio disable sndio support [autodetect] --disable-schannel disable SChannel SSP, needed for TLS support on Windows if openssl and gnutls are not used [autodetect] @@ -1799,6 +1800,7 @@ EXTERNAL_LIBRARY_LIST=" mediacodec openal opengl + pocketsphinx vapoursynth " @@ -3400,6 +3402,7 @@ afir_filter_deps="avcodec" afir_filter_select="fft" amovie_filter_deps="avcodec avformat" aresample_filter_deps="swresample" +asr_filter_deps="pocketsphinx" ass_filter_deps="libass" atempo_filter_deps="avcodec" atempo_filter_select="rdft" @@ -6298,6 +6301,7 @@ enabled openssl && { check_pkg_config openssl openssl openssl/ssl.h OP check_lib openssl openssl/ssl.h SSL_library_init -lssl32 -leay32 || check_lib openssl openssl/ssl.h SSL_library_init -lssl -lcrypto -lws2_32 -lgdi32 || die "ERROR: openssl not found"; } +enabled pocketsphinx && require_pkg_config pocketsphinx pocketsphinx pocketsphinx/pocketsphinx.h ps_init enabled rkmpp && { require_pkg_config rkmpp rockchip_mpp rockchip/rk_mpi.h mpp_create && require_pkg_config rockchip_mpp "rockchip_mpp >= 1.3.7" rockchip/rk_mpi.h mpp_create && { enabled libdrm || diff --git a/doc/filters.texi b/doc/filters.texi index 697ffcb921..38c70bf674 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -2131,6 +2131,41 @@ It accepts the following values: Set additional parameter which controls sigmoid function. @end table +@section asr +Automatic Speech Recognition + +This filter uses PocketSphinx for speech recognition. To enable +compilation of this filter, you need to configure FFmpeg with +@code{--enable-pocketsphinx}. + +It accepts the following options: + +@table @option +@item rate +Set sampling rate of input audio. Defaults is @code{16000}. +This need to match speech models, otherwise one will get poor results. + +@item hmm +Set dictionary containing acoustic model files. + +@item dict +Set pronunciation dictionary. + +@item lm +Set language model file. + +@item lmctl +Set language model set. + +@item lmname +Set which language model to use. + +@item logfn +Set output for log messages. +@end table + +The filter exports recognized speech as the frame metadata @code{lavfi.asr.text}. + @anchor{astats} @section astats diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 4ffa164345..b9109cc0d3 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -82,6 +82,7 @@ OBJS-$(CONFIG_ASHOWINFO_FILTER) += af_ashowinfo.o OBJS-$(CONFIG_ASIDEDATA_FILTER) += f_sidedata.o OBJS-$(CONFIG_ASOFTCLIP_FILTER) += af_asoftclip.o OBJS-$(CONFIG_ASPLIT_FILTER) += split.o +OBJS-$(CONFIG_ASR_FILTER) += af_asr.o OBJS-$(CONFIG_ASTATS_FILTER) += af_astats.o OBJS-$(CONFIG_ASTREAMSELECT_FILTER) += f_streamselect.o framesync.o OBJS-$(CONFIG_ATEMPO_FILTER) += af_atempo.o diff --git a/libavfilter/af_asr.c b/libavfilter/af_asr.c new file mode 100644 index 0000000000..0c08df1356 --- /dev/null +++ b/libavfilter/af_asr.c @@ -0,0 +1,181 @@ +/* + * Copyright (c) 2019 Paul B Mahol + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include + +#include "libavutil/avassert.h" +#include "libavutil/avstring.h" +#include "libavutil/channel_layout.h" +#include "libavutil/opt.h" +#include "audio.h" +#include "avfilter.h" +#include "internal.h" + +typedef struct ASRContext { + const AVClass *class; + + int rate; + char *hmm; + char *dict; + char *lm; + char *lmctl; + char *lmname; + char *logfn; + + ps_decoder_t *ps; + cmd_ln_t *config; + + int utt_started; +} ASRContext; + +#define OFFSET(x) offsetof(ASRContext, x) +#define FLAGS AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_FILTERING_PARAM +static const AVOption asr_options[] = { + { "rate", "set sampling rate", OFFSET(rate), AV_OPT_TYPE_INT, {.i64=16000}, 0, INT_MAX, .flags = FLAGS }, + { "hmm", "set directory containing acoustic model files", OFFSET(hmm), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, + { "dict", "set pronunciation dictionary", OFFSET(dict), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, + { "lm", "set language model file", OFFSET(lm), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, + { "lmctl", "set language model set", OFFSET(lmctl), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, + { "lmname","set which language model to use", OFFSET(lmname), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, + { "logfn", "set output for log messages", OFFSET(logfn), AV_OPT_TYPE_STRING, {.str="/dev/null"}, .flags = FLAGS }, + { NULL } +}; + +AVFILTER_DEFINE_CLASS(asr); + +static int filter_frame(AVFilterLink *inlink, AVFrame *in) +{ + AVFilterContext *ctx = inlink->dst; + AVDictionary **metadata = &in->metadata; + ASRContext *s = ctx->priv; + int have_speech; + const char *speech; + + ps_process_raw(s->ps, (const int16_t *)in->data[0], in->nb_samples, 0, 0); + have_speech = ps_get_in_speech(s->ps); + if (have_speech && !s->utt_started) + s->utt_started = 1; + if (!have_speech && s->utt_started) { + ps_end_utt(s->ps); + speech = ps_get_hyp(s->ps, NULL); + if (speech != NULL) + av_dict_set(metadata, "lavfi.asr.text", speech, 0); + ps_start_utt(s->ps); + s->utt_started = 0; + } + + return ff_filter_frame(ctx->outputs[0], in); +} + +static int config_input(AVFilterLink *inlink) +{ + AVFilterContext *ctx = inlink->dst; + ASRContext *s = ctx->priv; + + ps_start_utt(s->ps); + + return 0; +} + +static av_cold int asr_init(AVFilterContext *ctx) +{ + ASRContext *s = ctx->priv; + const float frate = s->rate; + char *rate = av_asprintf("%f", frate); + const char *argv[] = { "-logfn", s->logfn, + "-hmm", s->hmm, + "-lm", s->lm, + "-lmctl", s->lmctl, + "-lmname", s->lmname, + "-dict", s->dict, + "-samprate", rate, + NULL }; + + s->config = cmd_ln_parse_r(NULL, ps_args(), 14, (char **)argv, 0); + av_free(rate); + if (!s->config) + return AVERROR(ENOMEM); + + ps_default_search_args(s->config); + s->ps = ps_init(s->config); + if (!s->ps) + return AVERROR(ENOMEM); + + return 0; +} + +static int query_formats(AVFilterContext *ctx) +{ + ASRContext *s = ctx->priv; + int sample_rates[] = { s->rate, -1 }; + int ret; + + AVFilterFormats *formats = NULL; + AVFilterChannelLayouts *layout = NULL; + + if ((ret = ff_add_format (&formats, AV_SAMPLE_FMT_S16 )) < 0 || + (ret = ff_set_common_formats (ctx , formats )) < 0 || + (ret = ff_add_channel_layout (&layout , AV_CH_LAYOUT_MONO )) < 0 || + (ret = ff_set_common_channel_layouts (ctx , layout )) < 0 || + (ret = ff_set_common_samplerates (ctx , ff_make_format_list(sample_rates) )) < 0) + return ret; + + return 0; +} + +static av_cold void asr_uninit(AVFilterContext *ctx) +{ + ASRContext *s = ctx->priv; + + ps_free(s->ps); + s->ps = NULL; + cmd_ln_free_r(s->config); + s->config = NULL; +} + +static const AVFilterPad asr_inputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_AUDIO, + .filter_frame = filter_frame, + .config_props = config_input, + }, + { NULL } +}; + +static const AVFilterPad asr_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_AUDIO, + }, + { NULL } +}; + +AVFilter ff_af_asr = { + .name = "asr", + .description = NULL_IF_CONFIG_SMALL("Automatic Speech Recognition."), + .priv_size = sizeof(ASRContext), + .priv_class = &asr_class, + .init = asr_init, + .uninit = asr_uninit, + .query_formats = query_formats, + .inputs = asr_inputs, + .outputs = asr_outputs, +}; diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index 42f25522bd..78ed1cf804 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -74,6 +74,7 @@ extern AVFilter ff_af_ashowinfo; extern AVFilter ff_af_asidedata; extern AVFilter ff_af_asoftclip; extern AVFilter ff_af_asplit; +extern AVFilter ff_af_asr; extern AVFilter ff_af_astats; extern AVFilter ff_af_astreamselect; extern AVFilter ff_af_atempo; diff --git a/libavfilter/version.h b/libavfilter/version.h index 88287d940b..293af36f62 100644 --- a/libavfilter/version.h +++ b/libavfilter/version.h @@ -30,7 +30,7 @@ #include "libavutil/version.h" #define LIBAVFILTER_VERSION_MAJOR 7 -#define LIBAVFILTER_VERSION_MINOR 52 +#define LIBAVFILTER_VERSION_MINOR 53 #define LIBAVFILTER_VERSION_MICRO 100 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \