Smooth filter transitions

This adds per-sample smoothing of filter cutoff targets to remove zipper
noise during live tweaks. It also resets filter state to the requested
cutoff when notes start so attacks stay consistent.

Author: jserv

include/picosynth.h

@@ -90,9 +90,10 @@ typedef struct {
 
 /* Single-pole filter state */
 typedef struct {
-    const q15_t *in; /* Input signal pointer */
-    int32_t accum;   /* Internal accumulator (Q31) */
-    q15_t coeff;     /* Cutoff: 0=DC, Q15_MAX=bypass */
+    const q15_t *in;    /* Input signal pointer */
+    int32_t accum;      /* Internal accumulator (Q31) */
+    q15_t coeff;        /* Smoothed cutoff: 0=DC, Q15_MAX=bypass */
+    q15_t coeff_target; /* Target cutoff for smoothing */
 } picosynth_filter_t;
 
 /* 3-input mixer state */
@@ -189,6 +190,9 @@ void picosynth_init_hp(picosynth_node_t *n,
                        const q15_t *in,
                        q15_t coeff);
 
+/* Set filter cutoff with smoothing */
+void picosynth_filter_set_coeff(picosynth_node_t *n, q15_t coeff);
+
 /* Initialize 3-input mixer node */
 void picosynth_init_mix(picosynth_node_t *n,
                         const q15_t *gain,

src/picosynth.c

@@ -44,8 +44,10 @@ static void voice_note_on(picosynth_voice_t *v, uint8_t note)
         n->state = 0;
         n->out = 0;
         /* Reset filter accumulators to prevent DC offsets/pops */
-        if (n->type == PICOSYNTH_NODE_LP || n->type == PICOSYNTH_NODE_HP)
+        if (n->type == PICOSYNTH_NODE_LP || n->type == PICOSYNTH_NODE_HP) {
             n->flt.accum = 0;
+            n->flt.coeff = n->flt.coeff_target;
+        }
         /* Reset envelope block state to force immediate rate calculation */
         if (n->type == PICOSYNTH_NODE_ENV) {
             n->env.block_counter = 0;
@@ -325,6 +327,7 @@ void picosynth_init_lp(picosynth_node_t *n,
     n->type = PICOSYNTH_NODE_LP;
     n->flt.in = in;
     n->flt.coeff = coeff;
+    n->flt.coeff_target = coeff;
 }
 
 void picosynth_init_hp(picosynth_node_t *n,
@@ -337,6 +340,14 @@ void picosynth_init_hp(picosynth_node_t *n,
     n->type = PICOSYNTH_NODE_HP;
     n->flt.in = in;
     n->flt.coeff = coeff;
+    n->flt.coeff_target = coeff;
+}
+
+void picosynth_filter_set_coeff(picosynth_node_t *n, q15_t coeff)
+{
+    if (!n || (n->type != PICOSYNTH_NODE_LP && n->type != PICOSYNTH_NODE_HP))
+        return;
+    n->flt.coeff_target = q15_sat(coeff);
 }
 
 void picosynth_init_mix(picosynth_node_t *n,
@@ -504,6 +515,18 @@ q15_t picosynth_process(picosynth_t *s)
             }
             case PICOSYNTH_NODE_LP:
             case PICOSYNTH_NODE_HP: {
+                /* Smooth cutoff changes to avoid zipper noise (~4ms time
+                 * constant: delta/256 per sample).
+                 */
+                int32_t coeff_delta =
+                    (int32_t) n->flt.coeff_target - n->flt.coeff;
+                if (coeff_delta) {
+                    int32_t step = coeff_delta >> 8;
+                    if (step == 0)
+                        step = coeff_delta > 0 ? 1 : -1;
+                    n->flt.coeff = q15_sat((int32_t) n->flt.coeff + step);
+                }
+
                 /* Single-pole filter accumulator update:
                  * accum += (input - output)
                  * where output is the filtered signal from the previous sample.

web/wasm.c

@@ -182,7 +182,18 @@ void picosynth_wasm_set_filter_coeff(uint8_t voice, q15_t coeff)
         picosynth_params.v0_filter_coeff = coeff;
     else
         picosynth_params.v1_filter_coeff = coeff;
-    reinit_voices();
+
+    if (!picosynth_instance)
+        return;
+
+    picosynth_voice_t *v = picosynth_get_voice(picosynth_instance, voice);
+    if (!v)
+        return;
+
+    picosynth_node_t *flt = picosynth_voice_get_node(v, 0);
+    if (flt &&
+        (flt->type == PICOSYNTH_NODE_LP || flt->type == PICOSYNTH_NODE_HP))
+        picosynth_filter_set_coeff(flt, coeff);
 }
 
 EMSCRIPTEN_KEEPALIVE