diff options
| -rw-r--r-- | Android.bp | 17 | ||||
| -rw-r--r-- | METADATA | 12 | ||||
| -rw-r--r-- | README | 6 | ||||
| -rw-r--r-- | TRAINING-README (renamed from TRAINING) | 0 | ||||
| -rw-r--r-- | examples/rnnoise_demo.c | 4 | ||||
| -rw-r--r-- | include/rnnoise.h | 49 | ||||
| -rw-r--r-- | src/celt_lpc.c | 2 | ||||
| -rw-r--r-- | src/denoise.c | 4 | ||||
| -rw-r--r-- | src/kiss_fft.c | 2 | ||||
| -rw-r--r-- | src/rnn.h | 2 |
10 files changed, 87 insertions, 11 deletions
@@ -1,3 +1,20 @@ +package { + default_applicable_licenses: ["external_rnnoise_license"], +} + +// Added automatically by a large-scale-change +// See: http://go/android-license-faq +license { + name: "external_rnnoise_license", + visibility: [":__subpackages__"], + license_kinds: [ + "SPDX-license-identifier-BSD", + ], + license_text: [ + "COPYING", + ], +} + cc_library_static { name: "rnnoise_rnn_vad", // vendor needed for libpreprocessing effects. @@ -1,7 +1,5 @@ name: "rnnoise" -description: - "Recurrent neural network for audio noise reduction" - +description: "Recurrent neural network for audio noise reduction" third_party { url { type: HOMEPAGE @@ -11,7 +9,11 @@ third_party { type: GIT value: "https://github.com/xiph/rnnoise.git" } - version: "9acc1e5a633e0961a5895a73204df24744f199b6" - last_upgrade_date { year: 2020 month: 6 day: 3 } + version: "1cbdbcf1283499bbb2230a6b0f126eb9b236defd" license_type: NOTICE + last_upgrade_date { + year: 2021 + month: 2 + day: 9 + } } @@ -12,6 +12,10 @@ While it is meant to be used as a library, a simple command-line tool is provided as an example. It operates on RAW 16-bit (machine endian) mono PCM files sampled at 48 kHz. It can be used as: -./examples/rnnoise_demo <number of channels> <maximum attenuation> < input.raw > output.raw +./examples/rnnoise_demo <noisy speech> <output denoised> The output is also a 16-bit raw PCM file. + +The latest version of the source is available from +https://gitlab.xiph.org/xiph/rnnoise . The github repository +is a convenience copy. diff --git a/TRAINING b/TRAINING-README index 86c5a4e..86c5a4e 100644 --- a/TRAINING +++ b/TRAINING-README diff --git a/examples/rnnoise_demo.c b/examples/rnnoise_demo.c index 83d0709..c70343d 100644 --- a/examples/rnnoise_demo.c +++ b/examples/rnnoise_demo.c @@ -41,8 +41,8 @@ int main(int argc, char **argv) { fprintf(stderr, "usage: %s <noisy speech> <output denoised>\n", argv[0]); return 1; } - f1 = fopen(argv[1], "r"); - fout = fopen(argv[2], "w"); + f1 = fopen(argv[1], "rb"); + fout = fopen(argv[2], "wb"); while (1) { short tmp[FRAME_SIZE]; fread(tmp, sizeof(short), FRAME_SIZE, f1); diff --git a/include/rnnoise.h b/include/rnnoise.h index 67f0b06..c4215d9 100644 --- a/include/rnnoise.h +++ b/include/rnnoise.h @@ -30,6 +30,9 @@ #include <stdio.h> +#ifdef __cplusplus +extern "C" { +#endif #ifndef RNNOISE_EXPORT # if defined(WIN32) @@ -48,18 +51,64 @@ typedef struct DenoiseState DenoiseState; typedef struct RNNModel RNNModel; +/** + * Return the size of DenoiseState + */ RNNOISE_EXPORT int rnnoise_get_size(); +/** + * Return the number of samples processed by rnnoise_process_frame at a time + */ +RNNOISE_EXPORT int rnnoise_get_frame_size(); + +/** + * Initializes a pre-allocated DenoiseState + * + * If model is NULL the default model is used. + * + * See: rnnoise_create() and rnnoise_model_from_file() + */ RNNOISE_EXPORT int rnnoise_init(DenoiseState *st, RNNModel *model); +/** + * Allocate and initialize a DenoiseState + * + * If model is NULL the default model is used. + * + * The returned pointer MUST be freed with rnnoise_destroy(). + */ RNNOISE_EXPORT DenoiseState *rnnoise_create(RNNModel *model); +/** + * Free a DenoiseState produced by rnnoise_create. + * + * The optional custom model must be freed by rnnoise_model_free() after. + */ RNNOISE_EXPORT void rnnoise_destroy(DenoiseState *st); +/** + * Denoise a frame of samples + * + * in and out must be at least rnnoise_get_frame_size() large. + */ RNNOISE_EXPORT float rnnoise_process_frame(DenoiseState *st, float *out, const float *in); +/** + * Load a model from a file + * + * It must be deallocated with rnnoise_model_free() + */ RNNOISE_EXPORT RNNModel *rnnoise_model_from_file(FILE *f); +/** + * Free a custom model + * + * It must be called after all the DenoiseStates referring to it are freed. + */ RNNOISE_EXPORT void rnnoise_model_free(RNNModel *model); +#ifdef __cplusplus +} +#endif + #endif diff --git a/src/celt_lpc.c b/src/celt_lpc.c index 5d7ffa4..521351e 100644 --- a/src/celt_lpc.c +++ b/src/celt_lpc.c @@ -103,7 +103,7 @@ void celt_fir( { opus_val32 sum[4]; sum[0] = SHL32(EXTEND32(x[i ]), SIG_SHIFT); - sum[1] = SHL32(EXTEND32(x[i+1]), SIG_SHIFT), + sum[1] = SHL32(EXTEND32(x[i+1]), SIG_SHIFT); sum[2] = SHL32(EXTEND32(x[i+2]), SIG_SHIFT); sum[3] = SHL32(EXTEND32(x[i+3]), SIG_SHIFT); xcorr_kernel(rnum, x+i-ord, sum, ord); diff --git a/src/denoise.c b/src/denoise.c index d1c21dc..5a62844 100644 --- a/src/denoise.c +++ b/src/denoise.c @@ -257,6 +257,10 @@ int rnnoise_get_size() { return sizeof(DenoiseState); } +int rnnoise_get_frame_size() { + return FRAME_SIZE; +} + int rnnoise_init(DenoiseState *st, RNNModel *model) { memset(st, 0, sizeof(*st)); if (model) diff --git a/src/kiss_fft.c b/src/kiss_fft.c index 922dacc..d6b9f26 100644 --- a/src/kiss_fft.c +++ b/src/kiss_fft.c @@ -39,7 +39,7 @@ #define CUSTOM_MODES /* The guts header contains all the multiplication and addition macros that are defined for - complex numbers. It also delares the kf_ internal functions. + complex numbers. It also declares the kf_ internal functions. */ static void kf_bfly2( @@ -66,4 +66,4 @@ void compute_gru(const GRULayer *gru, float *state, const float *input); void compute_rnn(RNNState *rnn, float *gains, float *vad, const float *input); -#endif /* _MLP_H_ */ +#endif /* RNN_H_ */ |