private void ConfigureTimeRange(ref SpeakerResult currentSpeakerResult, Segment segment)
{
foreach (var item in segment.items)
{
if (currentSpeakerResult.end == 0m)
{
currentSpeakerResult.start = item.start_time;
}
currentSpeakerResult.end = item.end_time;
}
}
private async Task <SkillResult> ProcessTranscriptionResults(JObject transcriptionResults)
{
var result = new SkillResult();
StringBuilder speakerText = new StringBuilder();
TranscribeAlternative alternative = null;
var segments = transcriptionResults["results"]["speaker_labels"]["segments"].ToObject <List <Segment> >();
var transciptionsItems = transcriptionResults["results"]["items"].ToObject <List <TranscribeItem> >();
Console.WriteLine($"items: {transciptionsItems?.Count} segments: {segments.Count}");
var speakerLabel = string.Empty;
var lastSpeaker = "nobody";
SpeakerResult currentSpeakerResult = new SpeakerResult();
var itemIdx = 0;
var ti = transciptionsItems;
// sements have a begin and end, however the items contained in it also
// have begin and ends. the range of the items have a 1 to 1 correlation to the 'pronunciation' transcription
// item types. These also have ends which are outside the range of the segement strangely. So will be using segment to
// get the speaker, then will create an inclusive range for all items under it using the being of first and end of last.
foreach (var segment in segments)
{
if (segment.items.Length == 0)
{
continue;
}
result.duration = segment.end_time;
if (!lastSpeaker.Equals(segment.speaker_label))
{
// these lines do nothing the first iteration, but tie up last
// speaker result when the speaker is changing
currentSpeakerResult.text = speakerText.ToString();
speakerText = new StringBuilder();
// create new speaker result for new speaker - or first speaker on first iteration
var idx = result.speakerLabels.IndexOf(segment.speaker_label);
if (idx == -1)
{
idx = result.speakerLabels.Count;
result.speakerLabels.Add(segment.speaker_label);
result.resultBySpeaker.Add(idx, new List <SpeakerResult>());
}
currentSpeakerResult = new SpeakerResult();
currentSpeakerResult.speaker = idx;
ConfigureTimeRange(ref currentSpeakerResult, segment);
lastSpeaker = segment.speaker_label;
result.resultBySpeaker[idx].Add(currentSpeakerResult);
result.resultByTime.Add(currentSpeakerResult);
}
else
{
ConfigureTimeRange(ref currentSpeakerResult, segment);
}
for (; itemIdx < ti.Count &&
((currentSpeakerResult.start <= ti[itemIdx].start_time && ti[itemIdx].end_time <= currentSpeakerResult.end) ||
(ti[itemIdx].start_time == 0m))
; itemIdx++)
{
alternative = ti[itemIdx].alternatives.First();
if (alternative.content.Equals("[SILENCE]"))
{
speakerText.Append(".");
}
else
{
speakerText.Append(alternative.content);
}
speakerText.Append(" ");
}
}
currentSpeakerResult.text = speakerText.ToString();
// Call AWS Comprehend client to get sentiment for all speaker results
List <int> keyList = new List <int>(result.resultBySpeaker.Keys);
for (int keyIdx = 0; keyIdx < keyList.Count; keyIdx++)
{
var spkKey = keyList[keyIdx];
for (int resultIdx = result.resultBySpeaker[spkKey].Count - 1; resultIdx >= 0; resultIdx--)
{
if (!IsBlankText(result.resultBySpeaker[spkKey][resultIdx].text))
{
var speakerResult = result.resultBySpeaker[spkKey][resultIdx];
speakerResult.sentiment = await DetermineSentiment(result.resultBySpeaker[spkKey][resultIdx].text);
var topics = await DetermineTopic(result.resultBySpeaker[spkKey][resultIdx].text);
foreach (var topic in topics)
{
if (!result.topicLocations.ContainsKey(topic.Text))
{
result.topicLocations.Add(topic.Text, new List <SpeakerResult>());
}
result.topicLocations[topic.Text].Add(speakerResult);
}
}
}
}
return(result);
}
