Fixes 60 ms speech mode
[opus.git] / src / opus_encoder.c
index cf0eee1..bbbbcf4 100644 (file)
@@ -66,6 +66,7 @@ OpusEncoder *opus_encoder_create(int Fs, int channels)
 
     /* default SILK parameters */
     st->silk_mode.API_sampleRate        = st->Fs;
+    st->silk_mode.nChannels             = channels;
     st->silk_mode.maxInternalSampleRate = 16000;
     st->silk_mode.minInternalSampleRate = 8000;
     st->silk_mode.payloadSize_ms        = 20;
@@ -86,6 +87,7 @@ OpusEncoder *opus_encoder_create(int Fs, int channels)
        st->user_mode = OPUS_MODE_AUTO;
        st->user_bandwidth = BANDWIDTH_AUTO;
        st->voice_ratio = 90;
+       st->first = 1;
 
        st->encoder_buffer = st->Fs/100;
        st->delay_compensation = st->Fs/400;
@@ -109,7 +111,9 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
     int redundancy = 0;
     int redundancy_bytes = 0;
     int celt_to_silk = 0;
-    short pcm_buf[960*2];
+    /* TODO: This is 60 only so we can handle 60ms speech/audio switching 
+       it shouldn't bee too hard to reduce to 20 ms if needed */
+    short pcm_buf[3*960*2];
     int nb_compr_bytes;
     int to_celt = 0;
     celt_int32 mono_rate;
@@ -133,6 +137,9 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
     mono_rate = st->bitrate_bps;
     if (st->stream_channels==2)
         mono_rate = (mono_rate+10000)/2;
+    /* Compensate for smaller frame sizes assuming an equivalent overhead
+       of 60 bits/frame */
+    mono_rate -= 60*(st->Fs/frame_size - 50);
 
     /* Mode selection */
     if (st->user_mode==OPUS_MODE_AUTO)
@@ -154,32 +161,31 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
         st->mode = MODE_CELT_ONLY;
     }
 
-    /* FIXME: Remove this once SILK supports stereo */
-    if (st->channels == 2)
-        st->mode = MODE_CELT_ONLY;
-
     /* Bandwidth selection */
     if (st->mode == MODE_CELT_ONLY)
     {
-        if (mono_rate>35000 || (mono_rate>28000 && st->bandwidth==BANDWIDTH_FULLBAND))
-            st->bandwidth = BANDWIDTH_FULLBAND;
-        else if (mono_rate>28000 || (mono_rate>24000 && st->bandwidth==BANDWIDTH_SUPERWIDEBAND))
-            st->bandwidth = BANDWIDTH_SUPERWIDEBAND;
-        else if (mono_rate>24000 || (mono_rate>18000 && st->bandwidth==BANDWIDTH_WIDEBAND))
-            st->bandwidth = BANDWIDTH_WIDEBAND;
-        else
-            st->bandwidth = BANDWIDTH_NARROWBAND;
-    } else {
-        if (mono_rate>28000 || (mono_rate>24000 && st->bandwidth==BANDWIDTH_FULLBAND))
-            st->bandwidth = BANDWIDTH_FULLBAND;
-        else if (mono_rate>24000 || (mono_rate>18000 && st->bandwidth==BANDWIDTH_SUPERWIDEBAND))
-            st->bandwidth = BANDWIDTH_SUPERWIDEBAND;
-        else if (mono_rate>18000 || (mono_rate>14000 && st->bandwidth==BANDWIDTH_WIDEBAND))
-            st->bandwidth = BANDWIDTH_WIDEBAND;
-        else if (mono_rate>14000 || (mono_rate>11000 && st->bandwidth==BANDWIDTH_MEDIUMBAND))
-            st->bandwidth = BANDWIDTH_MEDIUMBAND;
-        else
-            st->bandwidth = BANDWIDTH_NARROWBAND;
+       if (mono_rate>35000 || (mono_rate>28000 && st->bandwidth==BANDWIDTH_FULLBAND))
+               st->bandwidth = BANDWIDTH_FULLBAND;
+       else if (mono_rate>28000 || (mono_rate>24000 && st->bandwidth==BANDWIDTH_SUPERWIDEBAND))
+               st->bandwidth = BANDWIDTH_SUPERWIDEBAND;
+       else if (mono_rate>24000 || (mono_rate>18000 && st->bandwidth==BANDWIDTH_WIDEBAND))
+               st->bandwidth = BANDWIDTH_WIDEBAND;
+       else
+               st->bandwidth = BANDWIDTH_NARROWBAND;
+    } else if (st->first || st->silk_mode.allowBandwidthSwitch)
+    {
+       if (mono_rate>30000 || (mono_rate>26000 && st->bandwidth==BANDWIDTH_FULLBAND))
+               st->bandwidth = BANDWIDTH_FULLBAND;
+       else if (mono_rate>22000 || (mono_rate>18000 && st->bandwidth==BANDWIDTH_SUPERWIDEBAND))
+               st->bandwidth = BANDWIDTH_SUPERWIDEBAND;
+       else if (mono_rate>16000 || (mono_rate>13000 && st->bandwidth==BANDWIDTH_WIDEBAND))
+               st->bandwidth = BANDWIDTH_WIDEBAND;
+       else if (mono_rate>13000 || (mono_rate>10000 && st->bandwidth==BANDWIDTH_MEDIUMBAND))
+               st->bandwidth = BANDWIDTH_MEDIUMBAND;
+       else
+               st->bandwidth = BANDWIDTH_NARROWBAND;
+       if (!st->silk_mode.inWBmodeWithoutVariableLP && st->bandwidth > BANDWIDTH_WIDEBAND)
+               st->bandwidth = BANDWIDTH_WIDEBAND;
     }
 
     if (st->Fs <= 24000 && st->bandwidth > BANDWIDTH_SUPERWIDEBAND)
@@ -266,12 +272,12 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
 
         st->silk_mode.payloadSize_ms = 1000 * frame_size / st->Fs;
         if (st->bandwidth == BANDWIDTH_NARROWBAND) {
-            st->silk_mode.maxInternalSampleRate = 8000;
+               st->silk_mode.desiredInternalSampleRate = 8000;
         } else if (st->bandwidth == BANDWIDTH_MEDIUMBAND) {
-            st->silk_mode.maxInternalSampleRate = 12000;
+               st->silk_mode.desiredInternalSampleRate = 12000;
         } else {
             SKP_assert( st->mode == MODE_HYBRID || st->bandwidth == BANDWIDTH_WIDEBAND );
-            st->silk_mode.maxInternalSampleRate = 16000;
+            st->silk_mode.desiredInternalSampleRate = 16000;
         }
         if( st->mode == MODE_HYBRID ) {
             /* Don't allow bandwidth reduction at lowest bitrates in hybrid mode */
@@ -279,6 +285,7 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
         } else {
             st->silk_mode.minInternalSampleRate = 8000;
         }
+        st->silk_mode.maxInternalSampleRate = 16000;
 
         /* Call SILK encoder for the low band */
         nBytes = max_data_bytes-1;
@@ -295,7 +302,6 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
         }
         if (nBytes==0)
             return 0;
-
         /* Extract SILK internal bandwidth for signaling in first byte */
         if( st->mode == MODE_SILK_ONLY ) {
             if( st->silk_mode.internalSampleRate == 8000 ) {
@@ -371,7 +377,6 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
             }
         }
 
-
         ec_enc_shrink(&enc, nb_compr_bytes);
        } else {
            nb_compr_bytes = 0;
@@ -487,6 +492,7 @@ int opus_encode(OpusEncoder *st, const short *pcm, int frame_size,
         st->prev_mode = MODE_CELT_ONLY;
     else
         st->prev_mode = st->mode;
+    st->first = 0;
     return ret+1+redundancy_bytes;
 }