diff --git a/nx2navi.ini.example b/nx2navi.ini.example
index c8fb3c95..2927848b 100644
--- a/nx2navi.ini.example
+++ b/nx2navi.ini.example
@@ -266,7 +266,7 @@
 [audio]
 # none mixes without opening an audio device.
 # audio = default
-# Audio the output queues before it starts, in ms (0 to 500).
+# Audio the output queues before it starts, in ms (0 to 500; above about 170, half the output ring, it is held there).
 # audio_prefill_ms = 20
 # Volume in percent (0 to 100), on top of the game's own. Changes while the game runs take effect at once.
 # volume = 100
diff --git a/runtime/audio_mixer.c b/runtime/audio_mixer.c
index 91822adf..21dc9db8 100644
--- a/runtime/audio_mixer.c
+++ b/runtime/audio_mixer.c
@@ -156,7 +156,9 @@ static void ring_pull(Mixer *m, float *out, uint32_t frames) {
     }
     uint64_t prefill = (uint64_t)m->frame * (uint64_t)m->prefill;
     if (prefill < frames) prefill = frames;
-    if (prefill > RING_FRAMES) prefill = RING_FRAMES;
+    // no more than the target the renderer fills the ring to (output_open caps it at half the ring): a longer
+    // prefill (audio_prefill_ms of 180 and up) was never reached, and the output stayed silent for good
+    if (prefill > RING_FRAMES / 2) prefill = RING_FRAMES / 2;
     if (!m->ready) {
         if (have < prefill) {
             memset(out, 0, (size_t)frames * oc * sizeof(float));
@@ -575,8 +577,12 @@ static void walk_step(Walk *w) {   /* the same transitions as advance_voice in a
     walk_skip_empty(w);
 }
 /* Decodes `frames` output samples per channel into out[c][k]; the renderer advances the voice afterwards by the same amount. */
+// Source samples per output sample. The guest's sample rate and pitch are not checked: an infinite or huge step
+// would keep the per-sample loops below (audio_voice_consumed, voice_render) running for good under the renderer's
+// lock, so it is held to 64 source samples per output sample, far above any pitch the game asks for.
 static double voice_step(const AudioRenderer *r, const ServerVoice *v) {
-    return v->in.sample_rate * (v->in.pitch > 0 ? v->in.pitch : 1.0) / r->param.sample_rate;
+    double step = v->in.sample_rate * (v->in.pitch > 0 ? v->in.pitch : 1.0) / r->param.sample_rate;
+    return step >= 0 && step <= 64 ? step : step > 64 ? 64 : 0;   // NaN: no progress
 }
 /* Source samples the voice consumes over `frames` output samples, counted with the exact accumulation voice_render steps by
    (float rounding makes `fraction + seconds * rate * pitch` differ from it by one sample on some frames, a seam click);
@@ -918,7 +924,8 @@ void audio_mixer_render(AudioRenderer *r, double seconds) {
             raw[2 * k] = s[0] * gain; raw[2 * k + 1] = s[1] * gain;   /* NX2NAVI_AUDIO_DUMP: the front pair before clipping */
             for (int c = 0; c < oc; c++) {
                 float x = s[c] * gain;
-                out[oc * k + c] = x > 1 ? 1 : x < -1 ? -1 : x;
+                // NaN (a NaN sample, an effect's state gone NaN) reaches the device as silence, not full scale
+                out[oc * k + c] = x > 1 ? 1 : x < -1 ? -1 : x == x ? x : 0;
             }
         }
         sink_dump(raw, frames);
diff --git a/runtime/audren.c b/runtime/audren.c
index a877febc..83f23f48 100644
--- a/runtime/audren.c
+++ b/runtime/audren.c
@@ -322,6 +322,8 @@ static void advance_voice(AudioRenderer *r, ServerVoice *v) {
     uint64_t samples = audio_voice_consumed(r, v, (int)r->param.sample_count);
     while (samples && v->valid[v->index]) {
         AudrenWaveBuffer *b = &v->buffers[v->index];
+        if (audio_buffer_empty(b)) break;   // an endless empty loop skip_empty kept: it starves the voice, and its
+                                            // zero length would make the loop below go round for good
         uint64_t length = (uint64_t)(b->end - b->start), left = length - v->offset;
         uint64_t step = samples < left ? samples : left;
         v->offset += step;