ai conversation (#49)

* capture audio with alsa

* add brainstorm thoughts

* use basecamp project for managing this project

* add libcurl basic example

* write pcm data to file

* transcribe audio file with deepgram

* transcribe raw audio

* hit anthropic API for ai question

* create full flow with stt, anthropic, tts, and playback

This organizes some components like deepgram into own module, and also
allows gets the full flow to work every time we run the program.

* organize audio and intelligence modules

* organize header files into include directory

* docs: add readme

* docs: explain future work

* docs: add disclaimer about hardcoded audio params

ai-conversation/include/audio.h

@@ -0,0 +1,29 @@
+#ifndef AUDIOCAPTURE_H
+#define AUDIOCAPTURE_H
+
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <alsa/asoundlib.h>
+#include <alsa/control.h>
+#include <alsa/pcm.h>
+
+// NOTE: find this via `arecord -l`
+#define ALSA_MIC_HARDWARE_DEVICE "hw:3"
+#define ALSA_SPEAKER_HARDWARE_DEVICE "default"
+
+// Captures audio for given seconds, and returns file name that we captured the
+// audio into
+char *audio_capture(uint seconds);
+
+/* count: how many bytes in the buffer
+ *  returns int: 0 if success, -1 if failure
+ *  NOTE: This function assumes that the audio data is using:
+ *  - linear16 formatting (each sample is 16 bits)
+ *  - one channel (1 sample)
+ *  - with 48000 HZ sampling rate
+ *  Hence, it's 2 bytes per frame.
+ */
+int audio_play(char *buf, size_t byte_count);
+
+#endif // AUDIOCAPTURE_H

ai-conversation/include/deepgram_client.h

@@ -0,0 +1,21 @@
+#ifndef DEEPGRAMCLIENT_H
+#define DEEPGRAMCLIENT_H
+
+#include <curl/curl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include "utils.h"
+
+/*
+ * @param audio_file: must already be opened
+ * @returns char *: transcript string result. Call must free.
+ */
+char *dg_transcribe(FILE *audio_file);
+
+/*
+*  @returns response data. Caller must free the struct AND the data within it.
+*  TODO: should return custom struct with details about encoding, sample rate, etc.
+*/
+struct curl_response_data *dg_text_to_speech(char *text);
+
+#endif // DEEPGRAMCLIENT_H

ai-conversation/include/intelligence.h

@@ -0,0 +1,9 @@
+#ifndef INTELLIGENCE_H
+#define INTELLIGENCE_H
+
+#include <stdio.h>
+#include <stdlib.h>
+
+char *ai_ask(char *question);
+
+#endif //INTELLIGENCE_H

ai-conversation/include/utils.h

@@ -0,0 +1,24 @@
+#ifndef UTILS_H
+#define UTILS_H
+
+#include <stdlib.h>
+#include <stdio.h>
+
+struct curl_file_data {
+  FILE *file;
+  size_t size;
+};
+
+struct curl_response_data {
+  char *data;
+  size_t size;
+};
+
+// This callback assumes the userdata is `struct file_data`
+size_t curl_read_callback_file(void *ptr, size_t size, size_t nmemb, void *userdata);
+
+// This callback assumes that we are aggregating callback data into an allocated response_data struct
+size_t curl_write_callback_response(void *contents, size_t size, size_t nmemb,
+                             void *userp);
+
+#endif // UTILS_H