This SDL2 / SDL2_ttf program records mono audio from any capture device, streams it to a WAV file, visualises the live waveform, and performs a lightweight autocorrelation‑based tuner that prints pitch (Hz) → musical note.

1. Initialise SDL (audio + video) and SDL_ttf.
2. Enumerate input devices and prompt user selection.
3. Open the chosen device (SDL_OpenAudioDevice) with floating‑point 48000 Hz.
4. Create an accelerated renderer & font for on‑screen text.
5. Pre‑write a WAV header to capture.wav.
6. Start capture – each callback fills a ring buffer.
7. Main loop:
   • Drain 2048 samples at a time, write them to disk,
   • detect pitch (detect_pitch) & translate to note (freq_to_note),
   • render scrolling waveform + pitch label.
8. On quit, finalise WAV sizes & release resources.

Settings block

#define WIN_W       1024   /* window width  */
#define WIN_H        400   /* window height */
#define BUF_SAMPLES  8192  /* ring‑buffer size (power‑of‑two) */
#define CHUNK        2048  /* processing hop size            */
#define WAV_NAME "capture.wav"
			

• Ring buffer: Two indices wr (writer) / rd (reader) wrap modulo BUF_SAMPLES.
• Tip: Changing CHUNK must remain > max lag (~960 samples) for reliable autocorrelation.

3.1 SDL / font init

SDL_Init(SDL_INIT_AUDIO | SDL_INIT_VIDEO) and TTF_Init() are mandatory before any I/O or rendering.

3.2 Device discovery

• SDL_GetNumAudioDevices(SDL_TRUE) counts capture devices.
• Names retrieved via SDL_GetAudioDeviceName.
• User choice read with scanf; defaults to 0.

3.3 Opening the stream

SDL_AudioSpec want = {0};
want.freq     = 48000;          /* sample‑rate */
want.format   = AUDIO_F32SYS;   /* 32‑bit float */
want.channels = 1;              /* mono */
want.samples  = CHUNK;          /* callback buffer size */
want.callback = capture_cb;     /* recording handler */
			

Returned have describes the exact format SDL could provide; the code uses it later for WAV header values.

3.4 Renderer & font

An SDL_Window → SDL_Renderer pair is set up.
Font path tries Apple SF first, falls back to DejaVu.

3.5 WAV header priming

Because the data chunk size is unknown upfront, the code back‑fills RIFF chunk size and data chunk size after recording stops (see Section 6).

3.6 Capture loop

• Poll SDL_QUIT events.
• When ring contains ≥ CHUNK samples:
  1. Copy to local[].
  2. Write raw floats to WAV.
  3. Compute pitch → note string.
  4. Render waveform polyline & text overlay.
• SDL_Delay(1) yields time to the OS.

capture_cb(void *udata, Uint8 *stream, int len) is invoked by SDL on a real‑time thread.

• stream → raw audio bytes (here float*).
• len     → byte count (÷ 4 for float samples).
• Function writes each sample into ring[wr], incrementing wr (wraps automatically via power‑of‑two mask).

⚠ Thread‑safety: Only wr is modified inside the callback; the main thread reads it without locks, relying on atomicity of 32‑bit writes.

float detect_pitch(const float *buf, int n, int sr)
			

• buf – pointer to CHUNK recent samples.
• n   – number of samples analysed.
• sr  – current sample‑rate.

For each candidate lag between minLag (1000 Hz) and maxLag (50 Hz) it performs a straight dot‑product of the signal with itself, retaining the lag with maximum energy. Returned pitch = sr / bestLag, or 0 if silence.

In practice this O(n·lag) scan is < 2 ms for 2048×900 floating operations on modern CPUs.

const char *freq_to_note(float f, char *out, size_t n)
			

• f   – detected frequency in Hz.
• out – pre‑allocated output buffer.
• n   – buffer capacity.

Uses the well‑known formula
  midi = 69 + 12·log2(f / 440 Hz)
then maps midi % 12 to name array {C, C#, … B} and puts octave = (midi/12 – 1).

After the loop exits, the code seeks back to two offsets:

1. Byte 4 (RIFF size) → 36 + bytes_total
2. data_size_pos → bytes_total

This finalises the file so any DAW/player can open it.
Finally all SDL objects, the device, and TTF subsystems are dequeued.

• Add SDL_AudioStream to resample mismatched devices.
• Maintain a scrolling trace (texture) rather than per‑frame redraw.
• Replace naive autocorrelation with YIN/AMDF for noisy inputs.
• Export additional metadata (timestamp, cent deviation).
• For stereo input, duplicate channels or run independent tuners.