1. 前端分析文本，用单词替换日期、数字和缩写，并发出描述发音的音素和韵律单元流。

2. 后端或合成器，它接受发音流并将其转换为声音。

• eSpeak 是 Ubuntu 和其他几个 Linux 发行版附带的 TTS 包。它是共振峰风格的，因此体积小（约 1.4MB），具有通常的机器人共振峰声音。如果需要，eSpeak 规范化器也可以与双音素合成器 (MBROLA) 一起使用，但我们不会在此处的清单示例中利用这一点。

• Flite 是 Festival 的嵌入式版本，Festival 是源自爱丁堡大学的开源语音合成包，Flite 是在卡内基梅隆大学完成的。它是基于双音素拼接的，正如您所期望的那样，它具有更自然的声音。CMU 还提供一套用于开发新声音的脚本和工具，称为 FestVox。

• Sphinx-4 是一个语音识别器和框架，可以使用多种识别方法，用 Java 编写。它主要用于服务器应用程序和研究。

• PocketSphinx 是一个从 Sphinx 派生的语音识别器，用 C 语言编写。因此，它比 Sphinx 小得多（但对于中等词汇量仍然约为 20MB），并且它可以在小型处理器上实时运行，即使是那些没有浮点硬件的处理器。

1. Freescale i.MX515 (ARM Cortex-A8 800MHz)。

2. 3D 图形处理单元。

3. WXGA 显示支持 (HDMI)。

4. 多格式高清视频解码器和 D1 视频编码器。

5. 512MB 内存。

6. 8GB 内部 SSD。

7. 10/100Mbit/s 以太网。

8. 802.11 b/g/n Wi-Fi。

9. SDHC 卡读卡器。

10. 2 个 USB 2.0 端口。

11. 用于耳机的音频插孔。

12. 内置扬声器。

清单 1. SafeSurgery.ckl

This is the Surgical Safety Checklist.
#
Before induction of anesthesia.
#
Has the patient confirmed his or her identity, site, 
procedure and consent?
# yes | no
Is the site marked?
# yes | notapplicable
Is the anesthesia machine and medication check complete?
# yes
Is the pulse oximeter on the patient and functioning?
# yes
Does the patient have a known allergy?
# yes | no
Does the patient have a difficult airway or aspiration risk?
# no | yes
Is there a risk of more than 500 milliliters of blood loss?
# no | yes
Thank you, that completes this portion of the checklist.

清单 2. atul.c

/*
   atul - a simple speech checklist for embedded systems
*/
#include <string.h>
#include <malloc.h>
#include <stdio.h>
#include <speak_lib.h>



espeak_POSITION_TYPE position_type;
espeak_AUDIO_OUTPUT output;
char *path=NULL;
int BuffLen=500, Options=0;
void* user_data;
t_espeak_callback *SynthCallback;
espeak_PARAMETER Parm;
FILE *ckfp;     /* Checklist file pointer */
char *ckBuf; /* Checklist item buffer */
char *mtchBuf; /* Checklist expected response buffer */
char *srBuf; /* Speech rec buffer */
char *reply; /* Trimmed reply */
int bsize=100; /* buffer length for all buffers */
int next;  /* flag - true if should go to next prompt */

char Voice[] = {"default"};
unsigned int size, position=0, end_position=0,
flags=espeakCHARS_AUTO|espeakENDPAUSE, *unique_identifier;

void recordreply(){
   /* read lines from stdin, which are piped in
    * from pocketsphinx_continuous.
    * Valid responses look like:
    * <9 digits>: reply (7 or 8 digits)
    * Returns a trimmed reply as char *reply
    * no spaces in return
    */
    int i, j;
    while (!feof(stdin)) {
       getline (&srBuf, &bsize, stdin);
       if (srBuf[9]!= ':') continue;
       j=0;
       for (i=0; i<strlen(srBuf); i++) {
          if (isdigit(srBuf[i])) continue;
          if (srBuf[i]=='-') continue;
          if (srBuf[i]==':') continue;
          if (isspace(srBuf[i])) continue;
          if (srBuf[i]=='(') continue;
          if (srBuf[i]==')') continue;
          reply[j++] = srBuf[i];
       }
       reply[j] = '\0';
       break;
    }
}

int checkreply(){
    /* returns true if reply matches
    *         false if no match (try again)
    */
   char *tryagain="Please answer ";
   char *ans, *spBuf;

   /* if template blank, just sleep 2 sec */
   if (strlen(mtchBuf)==2) {
      sleep(2);
      return(1);
   }
   /* see if reply matches template */
   recordreply();
   printf("reply: '%s'\n", reply);
   if (strstr(mtchBuf, reply)==NULL){
      /* no match - tell user what we're looking for */
      spBuf = (char *) malloc (bsize+1);
      strcpy(spBuf, tryagain);
      if (ans=strtok(mtchBuf+2, "|")){
         strcat(spBuf, ans);
      }
      ans = strtok(NULL, "|");
      while (ans!=NULL){
         strcat(spBuf, " or ");
         strcat(spBuf, ans);
         ans = strtok(NULL, "|");
      }
      espeak_Synth( spBuf, size, position,
         position_type, end_position, flags,
         unique_identifier, user_data );
      espeak_Synchronize( );
      free(spBuf);
      return(0); /* repeat last prompt */
   }else return(1); /* go to next prompt */
}

int main(int argc, char* argv[] )
{
    printf("atul started.\n");

    /* allocate needed buffers */
    ckBuf = (char *) malloc (bsize+1);
    srBuf = (char *) malloc (bsize+1);
    mtchBuf = (char *) malloc (bsize+1);
    reply = (char *) malloc (bsize+1);

    /* open the checklist file */
    if (argc < 2) {
        printf("Usage: atul <checklist filename>\n",
               argc);
        return 0;
    }
    ckfp = fopen(argv[1], "r");
    if (ckfp == NULL) {
        printf("Unable to open checklist file: %s\n",
               argv[1]);
        return 0;
    }

    /* Initialize the TTS subsystem */
    output = AUDIO_OUTPUT_PLAYBACK;
    espeak_Initialize(output, BuffLen, path, Options);
    espeak_SetVoiceByName(Voice);

    /* Initialize speech recognition
     *  piped in from pocketsphinx_continuous */
    while (!feof(stdin)) {
       getline (&srBuf, &bsize, stdin);
       if (strncmp(srBuf, "READY...", 8)==0) break;
    }

    /* Go through the checklist */
    next = 1;  /* advance to next prompt */
    while (!feof(ckfp)) {
       if (next) {
          getline (&ckBuf, &bsize, ckfp);
          getline (&mtchBuf, &bsize, ckfp);
       }
       size = strlen(ckBuf)+1;
       espeak_Synth( ckBuf, size, position,
          position_type, end_position, flags,
          unique_identifier, user_data );
          espeak_Synchronize( );
       fputs(ckBuf, stdout);
       next = checkreply();
    }
    fclose(ckfp);

    free(ckBuf);
    free(srBuf);
    free(mtchBuf);
    return 0;
}

资源

eSpeak: espeak.sourceforge.net/index.html

卡内基梅隆大学的 Sphinx 组: cmusphinx.sourceforge.net

卡内基梅隆大学的 Flite 页面: www.speech.cs.cmu.edu/flite

Genesi EFIKA MX: www.genesi-usa.com/products/efika

世界卫生组织安全手术清单: www.who.int/patientsafety/safesurgery/ss_checklist/en/index.html