Step 7.5b3a: per-source PCM converter (audio_out), shared by HLS and Spotify

- main/audio_out: one converter instance per source (own rate converter state): s16 any rate/channels -> 48 kHz stereo int32 -> ring. - player_write(src, ...) only writes for the active source, drops the rest; player_src_t is public in player.h. - HLS decoder uses audio_out (no behaviour change). - Verified: HLS still sample exact (441344 -> 480375 frames per 10.008 s segment), buffer ~4 s, 0 underruns. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-25 18:34:15 +10:00
parent c240ac9a3c
commit 5b0d7c9ba4
6 changed files with 131 additions and 73 deletions
@@ -0,0 +1,90 @@
+#include "audio_out.h"
+
+#include <stdlib.h>
+
+#include "esp_ae_rate_cvt.h"
+#include "esp_heap_caps.h"
+#include "esp_log.h"
+
+#define OUT_RATE        48000
+#define CONV_FRAMES     4096            // per rate converter call
+
+static const char *TAG = "audio_out";
+
+struct audio_conv {
+    player_src_t src;
+    const char *name;
+    esp_ae_rate_cvt_handle_t cvt;       // NULL: no conversion (source is 48 kHz)
+    uint32_t rate;                      // input rate the converter was opened for
+    int32_t *in32, *out32;              // stereo int32 work buffers (PSRAM)
+};
+
+audio_conv_t *audio_conv_create(player_src_t src, const char *name)
+{
+    audio_conv_t *c = calloc(1, sizeof(*c));
+    if (!c) {
+        return NULL;
+    }
+    c->src = src;
+    c->name = name;
+    c->in32 = heap_caps_malloc(CONV_FRAMES * 2 * sizeof(int32_t), MALLOC_CAP_SPIRAM);
+    c->out32 = heap_caps_malloc(CONV_FRAMES * 2 * 2 * sizeof(int32_t), MALLOC_CAP_SPIRAM);
+    if (!c->in32 || !c->out32) {
+        free(c->in32);
+        free(c->out32);
+        free(c);
+        return NULL;
+    }
+    return c;
+}
+
+static void set_rate(audio_conv_t *c, uint32_t rate)
+{
+    if (c->cvt) {
+        esp_ae_rate_cvt_close(c->cvt);
+        c->cvt = NULL;
+    }
+    c->rate = rate;
+    if (rate == OUT_RATE) {
+        return;
+    }
+    esp_ae_rate_cvt_cfg_t cfg = {
+        .src_rate = rate, .dest_rate = OUT_RATE, .channel = 2, .bits_per_sample = 32,
+        .complexity = 3, .perf_type = ESP_AE_RATE_CVT_PERF_TYPE_SPEED,
+    };
+    if (esp_ae_rate_cvt_open(&cfg, &c->cvt) != ESP_AE_ERR_OK) {
+        ESP_LOGE(TAG, "%s: rate converter %lu -> %d Hz: open failed", c->name, (unsigned long)rate, OUT_RATE);
+    } else {
+        ESP_LOGI(TAG, "%s: rate converter %lu -> %d Hz", c->name, (unsigned long)rate, OUT_RATE);
+    }
+}
+
+size_t audio_conv_write(audio_conv_t *c, const int16_t *pcm, uint32_t frames, int ch, uint32_t rate)
+{
+    if (rate != c->rate) {
+        set_rate(c, rate);
+    }
+    size_t produced = 0;
+    while (frames) {
+        uint32_t n = frames < CONV_FRAMES ? frames : CONV_FRAMES;
+        for (uint32_t i = 0; i < n; i++) {      // to stereo int32 (full scale = INT32_MAX)
+            int32_t l = (int32_t)pcm[i * ch] << 16;
+            c->in32[i * 2] = l;
+            c->in32[i * 2 + 1] = ch > 1 ? (int32_t)pcm[i * ch + 1] << 16 : l;
+        }
+        const int32_t *out = c->in32;
+        uint32_t out_n = n;
+        if (c->cvt) {
+            out_n = CONV_FRAMES * 2;
+            if (esp_ae_rate_cvt_process(c->cvt, c->in32, n, c->out32, &out_n) != ESP_AE_ERR_OK) {
+                ESP_LOGW(TAG, "%s: rate conversion failed", c->name);
+                return produced;
+            }
+            out = c->out32;
+        }
+        produced += player_write(c->src, out, out_n);
+        pcm += n * ch;
+        frames -= n;
+    }
+    return produced;
+}