feat: wip on implementing gemini analyzer

src/Directory.packages.props

@@ -7,6 +7,7 @@
     <PackageVersion Include="Microsoft.Extensions.Configuration" Version="9.0.7" />
     <PackageVersion Include="Microsoft.Extensions.Configuration.Json" Version="9.0.7" />
     <PackageVersion Include="Microsoft.Extensions.Hosting" Version="9.0.7" />
+    <PackageVersion Include="Microsoft.Extensions.Http" Version="9.0.7" />
     <PackageVersion Include="Microsoft.Extensions.Logging" Version="9.0.7" />
     <PackageVersion Include="NAudio" Version="2.2.1" />
     <PackageVersion Include="Serilog" Version="4.3.0" />

src/StreamShorts.Console/Commands/DefaultCommand.cs

@@ -4,13 +4,15 @@ internal sealed class DefaultCommand(
   IFileSystem fileSystem,
   IAnsiConsole console,
   IAudioExtractor audioExtractor,
-  ITranscriber transcriber
+  ITranscriber transcriber,
+  ITranscriptAnalyzer transcriptAnalyzer
 ) : AsyncCommand<DefaultCommand.Settings>
 {
   private readonly IFileSystem _fileSystem = fileSystem ?? throw new ArgumentNullException(nameof(fileSystem));
   private readonly IAnsiConsole _console = console ?? throw new ArgumentNullException(nameof(console));
   private readonly IAudioExtractor _audioExtractor = audioExtractor ?? throw new ArgumentNullException(nameof(audioExtractor));
   private readonly ITranscriber _transcriber = transcriber ?? throw new ArgumentNullException(nameof(transcriber));
+  private readonly ITranscriptAnalyzer _transcriptAnalyzer = transcriptAnalyzer ?? throw new ArgumentNullException(nameof(transcriptAnalyzer));
 
   internal class Settings : CommandSettings
   {
@@ -79,6 +81,15 @@ internal sealed class DefaultCommand(
 
     _console.MarkupLine($"[blue]Transcription completed[/] [green]successfully![/]");
 
+    TranscriptAnalysis? analysis = null;
+
+    await _console.Status()
+      .Spinner(Spinner.Known.Dots)
+      .StartAsync("Analyzing transcript...", async ctx =>
+      {
+        analysis = await _transcriptAnalyzer.AnalyzeAsync(transcriptionSegments);
+      });
+
     return 0;
   }
 }

src/StreamShorts.Console/Program.cs

@@ -38,89 +38,6 @@ finally
   await Log.CloseAndFlushAsync();
 }
 
-// // Step 3: Split the wave stream into 2 minute segments
-// waveStream.Position = 0;
-// var segmentDuration = TimeSpan.FromMinutes(2);
-// var segments = new List<MemoryStream>();
-// using var waveReader = new WaveFileReader(waveStream);
-// var segmentCount = (int)Math.Ceiling(waveReader.TotalTime.TotalMilliseconds / segmentDuration.TotalMilliseconds);
-
-// Directory.CreateDirectory("segments");
-
-// foreach (var i in Enumerable.Range(0, segmentCount))
-// {
-//   waveStream.Position = 0;
-//   using var segmentWaveReader = new WaveFileReader(waveStream);
-//   var segment = segmentWaveReader.ToSampleProvider()
-//     .Skip(i * segmentDuration)
-//     .Take(segmentDuration);
-//   var segmentProvider = segment.ToWaveProvider16();
-//   var segmentStream = new MemoryStream();
-//   WaveFileWriter.WriteWavFileToStream(segmentStream, segmentProvider);
-//   segmentStream.Position = 0;
-//   segments.Add(segmentStream);
-// }
-
-// // Step 4: Transcribe each segment using Whisper
-// using var modelMemoryStream = new MemoryStream();
-// var model = await WhisperGgmlDownloader.Default.GetGgmlModelAsync(GgmlType.TinyEn);
-// await model.CopyToAsync(modelMemoryStream);
-// var whisperFactory = WhisperFactory.FromBuffer(modelMemoryStream.ToArray());
-// using var whisperProcessor = whisperFactory.CreateBuilder()
-//   .WithLanguage("en")
-//   .Build();
-
-// var completeTranscription = new StringBuilder();
-
-// foreach (var (i, segment) in segments.Select((s, index) => (index, s)))
-// {
-//   var durationOffset = TimeSpan.FromMilliseconds(i * segmentDuration.TotalMilliseconds);
-//   var segmentTranscription = new StringBuilder();
-
-//   await foreach (var result in whisperProcessor.ProcessAsync(segment, CancellationToken.None))
-//   {
-//     var startTime = result.Start + durationOffset;
-//     var endTime = result.End + durationOffset;
-//     segmentTranscription.AppendLine(CultureInfo.CurrentCulture, $"[{startTime:hh\\:mm\\:ss} - {endTime:hh\\:mm\\:ss}] {result.Text}");
-//   }
-
-//   completeTranscription.Append(segmentTranscription);
-// }
-
-// // Step 5: Send the transcription to LLM for analysis
-// // TODO: Explore this prompt further...seems break
-// // when transcription is long
-// var prompt = $$"""
-// I need your help to transform my YouTube live stream transcript into engaging YouTube Shorts. Act as my content editor and pinpoint **all potential candidate segments** that are perfect for short-form video. I'm looking for clips that are:
-
-//   * **Funny:** Moments that will make viewers laugh.
-//   * **Informative:** Sections packed with valuable information or tips.
-//   * **Insightful:** Portions offering unique perspectives or 'aha\!' moments.
-
-// For each suggested short, please provide:
-
-//   * The **start time** of the initial segment and the **end time** of the final segment. The duration of each short should be no longer than 3 minutes, but **aim for durations between 15 seconds and 60 seconds**. However, the short **must be as long as necessary to capture the complete thought or idea**, even if it means exceeding the target range or extending slightly to capture all necessary dialogue.
-//   * A concise **title** that grabs attention.
-//   * A brief **description** highlighting the short's content and its appeal.
-//   * An **explanation** of why this particular segment is suitable for a YouTube Short, focusing on its potential for discoverability and engagement.
-
-// Please format your response as a JSON array of objects with the following structure:
-
-// ```json
-// {
-//   ""title"": ""string"",
-//   "start_time": "string",
-//   "end_time": "string",
-//   "description": "string",
-//   "explanation": "string"
-// }
-// ```
-
-// Here is the transcript of my YouTube live stream:
-
-// {{completeTranscription}}
-// """;
-
 // if (string.IsNullOrWhiteSpace(apiKey))
 // {
 //   throw new InvalidOperationException("GeminiApiKey is not configured in appsettings.json.");

src/StreamShorts.Console/Usings.cs

@@ -17,3 +17,4 @@ global using StreamShorts.Console.Commands;
 global using StreamShorts.Console.Hosting;
 global using StreamShorts.Library.Media.Audio;
 global using StreamShorts.Library.Transcription;
+global using StreamShorts.Library.Analysis;

src/StreamShorts.Library/Analysis/GeminiAnalyzer.cs

@@ -0,0 +1,88 @@
+using System.Text;
+using System.Text.Json;
+using System.Text.Json.Serialization;
+
+using NAudio.CoreAudioApi;
+
+using StreamShorts.Library.Transcription;
+
+namespace StreamShorts.Library.Analysis;
+
+public sealed class GeminiAnalyzer(
+  IHttpClientFactory httpClientFactory,
+  string apiKey
+) : ITranscriptAnalyzer
+{
+  private readonly IHttpClientFactory _httpClientFactory = httpClientFactory ?? throw new ArgumentNullException(nameof(httpClientFactory));
+  private readonly string _apiKey = apiKey ?? throw new ArgumentNullException(nameof(apiKey));
+
+  public async Task<TranscriptAnalysis> AnalyzeAsync(IEnumerable<TranscriptionSegment> segments)
+  {
+    using var client = _httpClientFactory.CreateClient();
+    client.Timeout = TimeSpan.FromMinutes(5);
+
+    // TODO: Load prompt from resource file
+    var requestUrl = $"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-lite:generateContent?key={_apiKey}";
+    using var request = new HttpRequestMessage(HttpMethod.Post, requestUrl)
+    {
+      Content = new StringContent(
+        JsonSerializer.Serialize(new
+        {
+          contents = new[]
+          {
+            new
+            {
+              role = "user",
+              parts = new[]
+              {
+                new
+                {
+                  text = prompt
+                }
+              },
+            }
+          },
+          generationConfig = new
+          {
+            responseMimeType = "application/json",
+          }
+        }),
+        Encoding.UTF8,
+        "application/json"
+      )
+    };
+    var response = await client.SendAsync(request).ConfigureAwait(false);
+    var responseContent = await response.Content.ReadAsStringAsync().ConfigureAwait(false);
+    var responseJson = JsonSerializer.Deserialize<LLMResponse>(responseContent);
+    var candidatesText = responseJson?
+      .Candidates?
+      .FirstOrDefault()?
+      .Content
+      .Parts?.FirstOrDefault()?
+      .Text;
+    
+    var clips = JsonSerializer.Deserialize<List<ShortClip>>(candidatesText ?? string.Empty);
+    return new TranscriptAnalysis(clips ?? []);
+  }
+}
+
+// TODO: Sort this shit out
+record LLMResponse(
+  [property: JsonPropertyName("candidates")]
+  Candidate[] Candidates
+);
+
+record Candidate(
+  [property: JsonPropertyName("content")]
+  Content Content
+);
+
+record Content(
+  [property: JsonPropertyName("parts")]
+  Part[] Parts
+);
+
+record Part(
+  [property: JsonPropertyName("text")]
+  string Text
+);

src/StreamShorts.Library/Analysis/ITranscriptAnalyzer.cs

@@ -0,0 +1,8 @@
+using StreamShorts.Library.Transcription;
+
+namespace StreamShorts.Library.Analysis;
+
+public interface ITranscriptAnalyzer
+{
+  Task<TranscriptAnalysis> AnalyzeAsync(IEnumerable<TranscriptionSegment> segments);
+}

src/StreamShorts.Library/Analysis/ShortClip.cs

@@ -0,0 +1,16 @@
+using System.Text.Json.Serialization;
+
+namespace StreamShorts.Library.Analysis;
+
+public record ShortClip(
+  [property: JsonPropertyName("title")]
+  string Title,
+  [property: JsonPropertyName("description")]
+  string Description,
+  [property: JsonPropertyName("explanation")]
+  string Explanation,
+  [property: JsonPropertyName("start_time")]
+  TimeSpan StartTime,
+  [property: JsonPropertyName("end_time")]
+  TimeSpan EndTime
+);

src/StreamShorts.Library/Analysis/TranscriptAnalysis.cs

@@ -0,0 +1,6 @@
+namespace StreamShorts.Library.Analysis;
+
+public class TranscriptAnalysis(IEnumerable<ShortClip> shortClips)
+{
+  public IEnumerable<ShortClip> ShortClips { get; init; } = shortClips;
+}

src/StreamShorts.Library/Resources/DefaultAnalysisPrompt.md

@@ -0,0 +1,28 @@
+I need your help to transform my YouTube live stream transcript into engaging YouTube Shorts. Act as my content editor and pinpoint **all potential candidate segments** that are perfect for short-form video. I'm looking for clips that are:
+
+  - **Funny:** Moments that will make viewers laugh.
+  - **Informative:** Sections packed with valuable information or tips.
+  - **Insightful:** Portions offering unique perspectives or 'aha\!' moments.
+
+For each suggested short, please provide:
+
+  - The **start time** of the initial segment and the **end time** of the final segment. The duration of each short should be no longer than 3 minutes, but **aim for durations between 15 seconds and 60 seconds**. However, the short **must be as long as necessary to capture the complete thought or idea**, even if it means exceeding the target range or extending slightly to capture all necessary dialogue.
+  - A concise **title** that grabs attention.
+  - A brief **description** highlighting the short's content and its appeal.
+  - An **explanation** of why this particular segment is suitable for a YouTube Short, focusing on its potential for discoverability and engagement.
+
+Please format your response as a JSON array of objects with the following structure:
+
+```json
+{
+  "title": "string",
+  "start_time": "string",
+  "end_time": "string",
+  "description": "string",
+  "explanation": "string"
+}
+```
+
+Here is the transcript of my YouTube live stream:
+
+{0}

src/StreamShorts.Library/StreamShorts.Library.csproj

@@ -1,6 +1,7 @@
 <Project Sdk="Microsoft.NET.Sdk">
   <ItemGroup>
     <PackageReference Include="FFMpegCore" />
+    <PackageReference Include="Microsoft.Extensions.Http" />
     <PackageReference Include="NAudio" />
     <PackageReference Include="Whisper.net.AllRuntimes" />
   </ItemGroup>