mirror of
https://github.com/rhasspy/espeak-ng.git
synced 2026-10-31 05:35:35 +00:00
Implement SpeechSynthesis.getChannelCount in Java.
This commit is contained in:
@@ -104,11 +104,6 @@ unicode_string::~unicode_string()
|
||||
#define LOG_TAG "eSpeakService"
|
||||
#define DEBUG true
|
||||
|
||||
enum audio_channel_count {
|
||||
CHANNEL_COUNT_MONO = 1,
|
||||
CHANNEL_COUNT_STEREO = 2
|
||||
};
|
||||
|
||||
enum audio_encoding {
|
||||
ENCODING_INVALID = 0x00,
|
||||
ENCODING_DEFAULT = 0x01,
|
||||
@@ -121,7 +116,6 @@ enum synthesis_result {
|
||||
SYNTH_ABORT = 1
|
||||
};
|
||||
|
||||
const int DEFAULT_CHANNEL_COUNT = CHANNEL_COUNT_MONO;
|
||||
const int DEFAULT_AUDIO_FORMAT = ENCODING_PCM_16BIT;
|
||||
const int DEFAULT_BUFFER_SIZE = 1000;
|
||||
|
||||
@@ -129,7 +123,6 @@ struct native_data_t {
|
||||
JNIEnv *env;
|
||||
jobject object;
|
||||
int sampleRate;
|
||||
int channelCount;
|
||||
int audioFormat;
|
||||
int bufferSizeInMillis;
|
||||
|
||||
@@ -137,7 +130,6 @@ struct native_data_t {
|
||||
env = NULL;
|
||||
object = NULL;
|
||||
sampleRate = 0;
|
||||
channelCount = DEFAULT_CHANNEL_COUNT;
|
||||
audioFormat = DEFAULT_AUDIO_FORMAT;
|
||||
bufferSizeInMillis = DEFAULT_BUFFER_SIZE;
|
||||
}
|
||||
@@ -247,14 +239,6 @@ JNICALL Java_com_reecedunn_espeak_SpeechSynthesis_nativeGetSampleRate(
|
||||
return (jint)(nat ? nat->sampleRate : 0);
|
||||
}
|
||||
|
||||
JNIEXPORT jint
|
||||
JNICALL Java_com_reecedunn_espeak_SpeechSynthesis_nativeGetChannelCount(
|
||||
JNIEnv *env, jobject object) {
|
||||
if (DEBUG) LOGV("%s", __FUNCTION__);
|
||||
const native_data_t *nat = getNativeData(env, object);
|
||||
return (jint) nat->channelCount;
|
||||
}
|
||||
|
||||
JNIEXPORT jint
|
||||
JNICALL Java_com_reecedunn_espeak_SpeechSynthesis_nativeGetAudioFormat(
|
||||
JNIEnv *env, jobject object) {
|
||||
|
||||
@@ -49,6 +49,8 @@ public class SpeechSynthesis {
|
||||
public static final int AGE_YOUNG = 12;
|
||||
public static final int AGE_OLD = 60;
|
||||
|
||||
public static final int CHANNEL_COUNT_MONO = 1;
|
||||
|
||||
static {
|
||||
System.loadLibrary("ttsespeak");
|
||||
|
||||
@@ -96,7 +98,7 @@ public class SpeechSynthesis {
|
||||
}
|
||||
|
||||
public int getChannelCount() {
|
||||
return nativeGetChannelCount();
|
||||
return CHANNEL_COUNT_MONO;
|
||||
}
|
||||
|
||||
public int getAudioFormat() {
|
||||
@@ -341,8 +343,6 @@ public class SpeechSynthesis {
|
||||
|
||||
private native final int nativeGetSampleRate();
|
||||
|
||||
private native final int nativeGetChannelCount();
|
||||
|
||||
private native final int nativeGetAudioFormat();
|
||||
|
||||
private native final int nativeGetBufferSizeInMillis();
|
||||
|
||||
Reference in New Issue
Block a user