diff --git a/hardware/alsa/api.c b/hardware/alsa/api.c
index dc762e06..a8806fc2 100644
--- a/hardware/alsa/api.c
+++ b/hardware/alsa/api.c
@@ -209,9 +209,11 @@ static int hw_dev_acquisition_start(const struct sr_dev_inst *sdi,
 	struct sr_datafeed_meta_analog meta;
 	struct dev_context *devc;
 	int count, ret;
+	char *endianess;
 
 	devc = sdi->priv;
 	devc->cb_data = cb_data;
+	devc->num_samples = 0;
 
 	sr_dbg("Setting audio access type to RW/interleaved.");
 	ret = snd_pcm_hw_params_set_access(devc->capture_handle,
@@ -222,9 +224,14 @@ static int hw_dev_acquisition_start(const struct sr_dev_inst *sdi,
 	}
 
 	/* FIXME: Hardcoded for 16bits. */
-	sr_dbg("Setting audio sample format to signed 16bit (little endian).");
+	if (SND_PCM_FORMAT_S16 == SND_PCM_FORMAT_S16_LE)
+		endianess = "lilltle endian";
+	else
+		endianess = "big endian";
+	sr_dbg("Setting audio sample format to signed 16bit (%s).", endianess);
 	ret = snd_pcm_hw_params_set_format(devc->capture_handle,
-			devc->hw_params, SND_PCM_FORMAT_S16_LE);
+					   devc->hw_params,
+					   SND_PCM_FORMAT_S16);
 	if (ret < 0) {
 		sr_err("Can't set audio sample format: %s.", snd_strerror(ret));
 		return SR_ERR;
diff --git a/hardware/alsa/protocol.c b/hardware/alsa/protocol.c
index 1c1d8849..a501ec8f 100644
--- a/hardware/alsa/protocol.c
+++ b/hardware/alsa/protocol.c
@@ -230,9 +230,10 @@ SR_PRIV int alsa_receive_data(int fd, int revents, void *cb_data)
 	struct dev_context *devc;
 	struct sr_datafeed_packet packet;
 	struct sr_datafeed_analog analog;
-	char inbuf[4096];
+	int16_t inbuf[4096];
 	int i, x, count, offset, samples_to_get;
-	uint16_t tmp16;
+	int16_t tmp16;
+	const float s16norm = 1 / (float)(1<<15);
 
 	(void)fd;
 	(void)revents;
@@ -262,11 +263,18 @@ SR_PRIV int alsa_receive_data(int fd, int revents, void *cb_data)
 	}
 
 	offset = 0;
-
-	for (i = 0; i < count; i++) {
+	/*
+	 * It's impossible to know what voltage levels the soundcard handles.
+	 * Some handle 0 dBV rms, some 0dBV peak-to-peak, +4dbmW (600 ohm), etc
+	 * Each of these corresponds to a different voltage, and there is no
+	 * mechanism to determine this voltage. The best solution is to send all
+	 * audio data as a normalized float, and let the frontend or user worry
+	 * about the calibration.
+	 */
+	for (i = 0; i < count; i += devc->num_probes) {
 		for (x = 0; x < devc->num_probes; x++) {
-			tmp16 = *(uint16_t *)(inbuf + (i * 4) + (x * 2));
-			analog.data[offset++] = (float)tmp16;
+			tmp16 = inbuf[i+x];
+			analog.data[offset++] = tmp16 * s16norm;
 		}
 	}