// === FixSpeakerTags method ===
// The LongRunningRecognizeResponse does not put SpeakerTag values
// on the words initially until it has completed the transcription.
// At that point, it creates one more result that has the entire text
// in its Transcript field, and a word array that contains
// every word in the entire text. The SpeakerTag fields now contains values.
// FixSpeakerTags moves the SpeakerTag values from the last result in the response
// to the corresponding words in the initial results and then removes the final result.
public static Transcribed_Dto FixSpeakerTags(Transcribed_Dto transcribed)
{
int resultCount = transcribed.Talks.Count;
TranscribedTalk_Dto lastResult = transcribed.Talks[resultCount - 1];
TranscribedTalk_Dto nextToLastResult = transcribed.Talks[resultCount - 2];
int lastWordnum = lastResult.Words[^ 1].WordNum;
/* === TransformResponse.Simplify method ===
* We want to extract all the useful data from the response that comes back from the cloud.
* But we don't want the superlous fields that make it more complicated to use.
*
* The raw response structure contains:
* A single unnamed object with a "Results" array.
* The "Results" array consists of unnamed objects, each containing:
* "Alternatives" array, "ChannelTag" integer, "LanguageCode" string
* The "Alternatives" arrays appear to always consists of a single unnamed object containing:
* "Transcript" string, "Confidence" decimal, "Words" array
* WHEN DOES THIS EVER CONSIST OF MORE THEN ONE ALTERNATIVE?
* The "Words" array consists of unnamed objects containing:
* "StartTime" object, "EndTime" object, "Word" object
* The "StartTime" and "EndTime" objects both contain:
* "Seconds" int, "Nanos" integer
* The "Word" objects contain:
* "Word" string, "Confidence" decimal, "SpeakerTag" integer
*
* The new structure contains:
* A single unnamed object with a "Results" array.
* The "Results" array consists of unnamed objects, each containing:
* "Transcript" string, "Confidence" decimal, "Words" array and "WordCount" integer
* The "Words" array consists of unnamed objects, eash containing:
* "Word" string, "Confidence" decimal, "StartTime" integer, "EndTime integer, "speakerTag" integer,
* and "WordNum" integer.
* Both StartTime and EndTime integers are in milliseconds.
* "WordCount" and "WordNum" are new fields added to help in fixing speaker tags,
* but we leave them in the final structure for possible future use.
*/
public static Transcribed_Dto Simpify(RepeatedField <SpeechRecognitionResult> recogResults)
{
Transcribed_Dto transcript = new Transcribed_Dto();
int altCount = 0;
int wordNum = 0;
foreach (SpeechRecognitionResult recogResult in recogResults)
{
if (recogResult.Alternatives.Count > 1)
{
altCount++;
Console.WriteLine($"ERROR: more than 1 alternative - result {altCount}");
}
;
SpeechRecognitionAlternative recogAlt = recogResult.Alternatives[0];
TranscribedTalk_Dto result = new TranscribedTalk_Dto(recogAlt.Transcript, recogAlt.Confidence)
{
// The new "WordCount" field in Result is populated with the total word count.
WordCount = recogAlt.Words.Count,
};
Console.WriteLine($"Next result: {recogAlt.Words.Count} words");
foreach (var item in recogAlt.Words)
{
long startTime = item.StartTime.Seconds * 1000 + item.StartTime.Nanos / 1000000;
long endTime = item.EndTime.Seconds * 1000 + item.EndTime.Nanos / 1000000;
// The new "WordNum" field in RespWord is popluated with the sequencial "wordnum"
wordNum++;
result.Words.Add(new TranscribedWord_Dto(item.Word, item.Confidence, startTime, endTime, item.SpeakerTag, wordNum));
}
transcript.Talks.Add(result);
}
return(transcript);
}
