]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/g729dec.c
free subtitle_header before overwriting it to avoid memleak
[ffmpeg] / libavcodec / g729dec.c
index c21e89715cdbd587528f6314b3776260c246d32c..3a6fb0fb7a2b6f1ade9d4da9bb022adc780ff443 100644 (file)
@@ -81,6 +81,13 @@ typedef struct {
 } G729FormatDescription;
 
 typedef struct {
+    int pitch_delay_int_prev;   ///< integer part of previous subframe's pitch delay (4.1.3)
+
+    /// (2.13) LSP quantizer outputs
+    int16_t  past_quantizer_output_buf[MA_NP + 1][10];
+    int16_t* past_quantizer_outputs[MA_NP + 1];
+
+    int16_t lsfq[10];           ///< (2.13) quantized LSF coefficients from previous frame
     int16_t lsp_buf[2][10];     ///< (0.15) LSP coefficients (previous and current frames) (3.2.5)
     int16_t *lsp[2];            ///< pointers to lsp_buf
 }  G729Context;
@@ -119,16 +126,63 @@ static inline int get_parity(uint8_t value)
    return (0x6996966996696996ULL >> (value >> 2)) & 1;
 }
 
+static void lsf_decode(int16_t* lsfq, int16_t* past_quantizer_outputs[MA_NP + 1],
+                       int16_t ma_predictor,
+                       int16_t vq_1st, int16_t vq_2nd_low, int16_t vq_2nd_high)
+{
+    int i,j;
+    static const uint8_t min_distance[2]={10, 5}; //(2.13)
+    int16_t* quantizer_output = past_quantizer_outputs[MA_NP];
+
+    for (i = 0; i < 5; i++) {
+        quantizer_output[i]     = cb_lsp_1st[vq_1st][i    ] + cb_lsp_2nd[vq_2nd_low ][i    ];
+        quantizer_output[i + 5] = cb_lsp_1st[vq_1st][i + 5] + cb_lsp_2nd[vq_2nd_high][i + 5];
+    }
+
+    for (j = 0; j < 2; j++) {
+        for (i = 1; i < 10; i++) {
+            int diff = (quantizer_output[i - 1] - quantizer_output[i] + min_distance[j]) >> 1;
+            if (diff > 0) {
+                quantizer_output[i - 1] -= diff;
+                quantizer_output[i    ] += diff;
+            }
+        }
+    }
+
+    for (i = 0; i < 10; i++) {
+        int sum = quantizer_output[i] * cb_ma_predictor_sum[ma_predictor][i];
+        for (j = 0; j < MA_NP; j++)
+            sum += past_quantizer_outputs[j][i] * cb_ma_predictor[ma_predictor][j][i];
+
+        lsfq[i] = sum >> 15;
+    }
+
+    /* Rotate past_quantizer_outputs. */
+    memmove(past_quantizer_outputs + 1, past_quantizer_outputs, MA_NP * sizeof(int16_t*));
+    past_quantizer_outputs[0] = quantizer_output;
+
+    ff_acelp_reorder_lsf(lsfq, LSFQ_DIFF_MIN, LSFQ_MIN, LSFQ_MAX, 10);
+}
+
 static av_cold int decoder_init(AVCodecContext * avctx)
 {
+    G729Context* ctx = avctx->priv_data;
+    int i,k;
+
     if (avctx->channels != 1) {
         av_log(avctx, AV_LOG_ERROR, "Only mono sound is supported (requested channels: %d).\n", avctx->channels);
-        return AVERROR_NOFMT;
+        return AVERROR(EINVAL);
     }
 
     /* Both 8kbit/s and 6.4kbit/s modes uses two subframes per frame. */
     avctx->frame_size = SUBFRAME_SIZE << 1;
 
+    for (k = 0; k < MA_NP + 1; k++) {
+        ctx->past_quantizer_outputs[k] = ctx->past_quantizer_output_buf[k];
+        for (i = 1; i < 11; i++)
+            ctx->past_quantizer_outputs[k][i - 1] = (18717 * i) >> 3;
+    }
+
     ctx->lsp[0] = ctx->lsp_buf[0];
     ctx->lsp[1] = ctx->lsp_buf[1];
     memcpy(ctx->lsp[0], lsp_init, 10 * sizeof(int16_t));
@@ -154,6 +208,9 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
     uint8_t quantizer_2nd_lo; ///< second stage lower vector of quantizer (size in bits)
     uint8_t quantizer_2nd_hi; ///< second stage higher vector of quantizer (size in bits)
 
+    int pitch_delay_int;         // pitch delay, integer part
+    int pitch_delay_3x;          // pitch delay, multiplied by 3
+
     if (*data_size < SUBFRAME_SIZE << 2) {
         av_log(avctx, AV_LOG_ERROR, "Error processing packet: output buffer too small\n");
         return AVERROR(EIO);
@@ -167,7 +224,7 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
         av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729D @ 6.4kbit/s");
     } else {
         av_log(avctx, AV_LOG_ERROR, "Packet size %d is unknown.\n", buf_size);
-        return (AVERROR_NOFMT);
+        return AVERROR_INVALIDDATA;
     }
 
     for (i=0; i < buf_size; i++)
@@ -181,6 +238,10 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
     quantizer_2nd_lo = get_bits(&gb, VQ_2ND_BITS);
     quantizer_2nd_hi = get_bits(&gb, VQ_2ND_BITS);
 
+    lsf_decode(ctx->lsfq, ctx->past_quantizer_outputs,
+               ma_predictor,
+               quantizer_1st, quantizer_2nd_lo, quantizer_2nd_hi);
+
     ff_acelp_lsf2lsp(ctx->lsp[1], ctx->lsfq, 10);
 
     ff_acelp_lp_decode(&lp[0][0], &lp[1][0], ctx->lsp[1], ctx->lsp[0], 10);
@@ -202,12 +263,30 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
         gc_1st_index  = get_bits(&gb, format.gc_1st_index_bits);
         gc_2nd_index  = get_bits(&gb, format.gc_2nd_index_bits);
 
-        ff_acelp_weighted_vector_sum(fc + pitch_delay_int[i],
-                                     fc + pitch_delay_int[i],
+        if(!i) {
+            if (bad_pitch)
+                pitch_delay_3x   = 3 * ctx->pitch_delay_int_prev;
+            else
+                pitch_delay_3x = ff_acelp_decode_8bit_to_1st_delay3(ac_index);
+        } else {
+            int pitch_delay_min = av_clip(ctx->pitch_delay_int_prev - 5,
+                                          PITCH_DELAY_MIN, PITCH_DELAY_MAX - 9);
+
+            if(packet_type == FORMAT_G729D_6K4)
+                pitch_delay_3x = ff_acelp_decode_4bit_to_2nd_delay3(ac_index, pitch_delay_min);
+            else
+                pitch_delay_3x = ff_acelp_decode_5_6_bit_to_2nd_delay3(ac_index, pitch_delay_min);
+        }
+
+        /* Round pitch delay to nearest (used everywhere except ff_acelp_interpolate). */
+        pitch_delay_int  = (pitch_delay_3x + 1) / 3;
+
+        ff_acelp_weighted_vector_sum(fc + pitch_delay_int,
+                                     fc + pitch_delay_int,
                                      fc, 1 << 14,
                                      av_clip(ctx->gain_pitch, SHARP_MIN, SHARP_MAX),
                                      0, 14,
-                                     SUBFRAME_SIZE - pitch_delay_int[i]);
+                                     SUBFRAME_SIZE - pitch_delay_int);
 
         if (frame_erasure) {
             ctx->gain_pitch = (29491 * ctx->gain_pitch) >> 15; // 0.90 (0.15)
@@ -225,6 +304,8 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
                                      (!voicing && frame_erasure) ? 0 : ctx->gain_pitch,
                                      ( voicing && frame_erasure) ? 0 : ctx->gain_code,
                                      1 << 13, 14, SUBFRAME_SIZE);
+
+            ctx->pitch_delay_int_prev = pitch_delay_int;
     }
 
     *data_size = SUBFRAME_SIZE << 2;
@@ -234,7 +315,7 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
 AVCodec g729_decoder =
 {
     "g729",
-    CODEC_TYPE_AUDIO,
+    AVMEDIA_TYPE_AUDIO,
     CODEC_ID_G729,
     sizeof(G729Context),
     decoder_init,