marking
/
Learn.VideoAnalysis
19
0
Fork 0
Code Pull Requests Actions Releases Activity

新增 封装vad,接入 FunASRNano

This commit is contained in:
小肥羊 2026-01-13 17:42:07 +08:00
parent de1cdcf32c
commit d52504a3a0
6 changed files with 375 additions and 176 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
VideoAnalysis/Program.cs
View File

@ -53,6 +53,7 @@ namespace Learn.VideoAnalysis
builder.Services.AddAlibabaCloudVod();
builder.Services.AddAliyunOSS();
builder.Services.AddSenseVoiceExpand();
builder.Services.AddSherpaVadExpand();
//builder.Services.AddSpeakerAI();
builder.Services.AddCoravel();

132
VideoAnalysisCore/AICore/SherpaOnnx/FunASRNano.cs Normal file
View File

@ -0,0 +1,132 @@
using Microsoft.Extensions.DependencyInjection;
using Microsoft.Extensions.Options;
using SherpaOnnx;
using SqlSugar.IOC;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.IO;
using System.Linq;
using System.Text;
using System.Text.Json;
using System.Text.RegularExpressions;
using System.Threading.Tasks;
using VideoAnalysisCore.Common;
using VideoAnalysisCore.Model;
using VideoAnalysisCore.Model.Enum;
namespace VideoAnalysisCore.AICore.SherpaOnnx
{
public static class FunASRNanoExpand
{
/// <summary>
/// 添加 SenseVoice 语音转文字
/// </summary>
/// <param name="services"></param>
public static void AddFunASRNanoExpand(this IServiceCollection services)
{
services.AddSingleton<SenseVoice>();
}
}
/// <summary>
/// 基于 sherpa-onnx 平台接入的 Fun-ASR-Nano-2512
/// <para>版本 Fun-ASR-Nano-2512</para>
/// <para>来源 https://github.com/modelscope/FunASR/blob/main/README_zh.md</para>
/// </summary>
public class FunASRNano
{
static OfflineRecognizer OR = default!;
private readonly IServiceProvider serviceProvider;
public FunASRNano( RedisManager redisManager, IServiceProvider serviceProvider)
{
this.serviceProvider = serviceProvider;
}
/// <summary>
/// 初始化 SenseVoice
/// </summary>
/// <param name="numThreads">默认6线程</param>
/// <param name="useGPU">是否使用gpu 报错请看安装CUDA环境 <see cref="https://k2-fsa.github.io/sherpa/onnx/pretrained_models/whisper/large-v3.html#run-with-gpu-float32"/></param>
public void Init(int numThreads = 6, bool useGPU = false, bool useHotwords = false)
{
Console.WriteLine("初始化 FunASRNano");
OfflineRecognizerConfig config = new OfflineRecognizerConfig();
//采样率
config.FeatConfig.SampleRate = 16000;
//用于训练模型的特征维度
config.FeatConfig.FeatureDim = 80;
var topFolder = Path.Combine(AppCommon.AIModelFile, "sherpa-onnx-funasr-nano-fp16-2025-12-30");
//模型配置
//将非结构化数据（文本、图像、音频等）转换为低维稠密向量
config.ModelConfig.FunAsrNano.EncoderAdaptor = Path.Combine(topFolder, "encoder_adaptor.int8.onnx");
//接入的大语言模型
config.ModelConfig.FunAsrNano.LLM = Path.Combine(topFolder, "llm.fp16.onnx");
//插入预训练模型（如Transformer）的小型可训练模块 (如语音识别、情感分析)
config.ModelConfig.FunAsrNano.Embedding = Path.Combine(topFolder, "embedding.int8.onnx");
//分词器
config.ModelConfig.FunAsrNano.Tokenizer = Path.Combine(topFolder, "Qwen3-0.6B");
//提示词
config.ModelConfig.FunAsrNano.SystemPrompt = "You are a professional video audio transcription assistant.";
config.ModelConfig.FunAsrNano.UserPrompt = "这是一趟中国的课堂视频音频,请你帮我分析出它讲述的内容!";
config.ModelConfig.FunAsrNano.MaxNewTokens = 512;
config.ModelConfig.FunAsrNano.Temperature = 1E-06f;
config.ModelConfig.FunAsrNano.TopP = 0.8f;
config.ModelConfig.FunAsrNano.Seed = 42;
//模型类型
config.ModelConfig.ModelType = string.Empty;
config.ModelConfig.NumThreads = numThreads;
config.ModelConfig.Provider = "cpu";
//需要使用GPU
if (!useGPU)
config.ModelConfig.Provider = "cuda";
#if DEBUG
config.ModelConfig.Debug = 1;
#endif
OR = new OfflineRecognizer(config);
}
/// <summary>
/// 获取语音字幕
/// </summary>
/// <param name="s"></param>
/// <returns></returns>
public List<SenseVoiceRes> RunTask(Stream s)
{
if (s is null) throw new Exception("音频路径 is null");
return serviceProvider.GetRequiredService<SherpaVad>()
.TaskHandle(new WaveReader(s), null, SoundHandle, SherpaVadVersion.silero_vad_v5);
}
/// <summary>
/// 获取语音字幕
/// </summary>
/// <param name="task"></param>
/// <returns></returns>
public Task RunTask(string task)
{
var filePath = Path.Combine(task.LocalPath(), "task.wav");
if (string.IsNullOrEmpty(filePath) || !File.Exists(filePath))
throw new Exception("task 音频路径未找到");
serviceProvider.GetRequiredService<SherpaVad>()
.TaskHandle(new WaveReader(filePath), null, SoundHandle, SherpaVadVersion.silero_vad_v5);
return Task.CompletedTask;
}
/// <summary>
/// 获取语音字幕
/// </summary>
/// <param name="sampleRate">采样率</param>
/// <param name="samples">采样值(样品)</param>
/// <returns>结果流</returns>
public OfflineStream SoundHandle(int sampleRate, float[] samples)
{
var stream = OR.CreateStream();
stream.AcceptWaveform(sampleRate, samples);
OR.Decode(stream);
return stream;
}
}
}

198
VideoAnalysisCore/AICore/SherpaOnnx/SenseVoice.cs
View File

@ -14,7 +14,6 @@ using System.Threading.Tasks;
using VideoAnalysisCore.Common;
using VideoAnalysisCore.Model;
using VideoAnalysisCore.Model.Enum;
using static System.Runtime.InteropServices.JavaScript.JSType;
namespace VideoAnalysisCore.AICore.SherpaOnnx
{
@ -32,22 +31,18 @@ namespace VideoAnalysisCore.AICore.SherpaOnnx
}
public class SenseVoice
{
//const string TransducerStr = "sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20";
static OfflineRecognizer OR = default!;
static OfflineRecognizer OR_old = default!;
static VadModelConfig VADModelConfig = default!;
public Repository<VideoTask> videoTaskDB { get; set; }
static OfflineRecognizer OR = default!;
private readonly RedisManager redisManager;
private readonly IServiceProvider serviceProvider;
public SenseVoice(Repository<VideoTask> videoTaskDB, RedisManager redisManager)
public SenseVoice(RedisManager redisManager, IServiceProvider serviceProvider)
{
this.videoTaskDB = videoTaskDB;
this.redisManager = redisManager;
this.serviceProvider = serviceProvider;
}
/// <summary>
/// 初始化 SenseVoice
/// 初始化 SenseVoice
/// </summary>
/// <param name="numThreads">默认6线程</param>
/// <param name="useGPU">是否使用gpu 报错请看安装CUDA环境<see cref="https://k2-fsa.github.io/sherpa/onnx/pretrained_models/whisper/large-v3.html#run-with-gpu-float32"/></param>
@ -61,10 +56,9 @@ namespace VideoAnalysisCore.AICore.SherpaOnnx
config.FeatConfig.FeatureDim = 80;
// Path to tokens.txt
var AIModelVersion_270717 = "sherpa-onnx-sense-voice-24-07-17";
var AIModelVersion_251217 = "sherpa-onnx-sense-voice-funasr-nano-2025-12-17";
config.ModelConfig.Tokens = Path.Combine(AppCommon.AIModelFile, AIModelVersion_251217, "tokens.txt");
config.ModelConfig.Tokens = Path.Combine(AppCommon.AIModelFile, AIModelVersion_270717, "tokens.txt");
//SenseVoice 模型
config.ModelConfig.SenseVoice.Model = Path.Combine(AppCommon.AIModelFile, AIModelVersion_251217, "model.onnx");
config.ModelConfig.SenseVoice.Model = Path.Combine(AppCommon.AIModelFile, AIModelVersion_270717, "model.onnx");
//1 使用逆文本规范化处理感官语音 [控制标点符号生成]。
config.ModelConfig.SenseVoice.UseInverseTextNormalization = 1;
//反转文本规范化规则 fst 的路径
@ -91,54 +85,11 @@ namespace VideoAnalysisCore.AICore.SherpaOnnx
//config.MaxActivePaths =4;
#endregion
#region 热词功能[无效]
//if (false)
//{
// //热词目录
// config.HotwordsFile = Path.Combine(AppCommon.AIModelFile, "Hotwords.txt");
// config.DecodingMethod = "modified_beam_search";
// //热词得分
// config.HotwordsScore = 1.5f;
// config.ModelConfig.ModelingUnit = "cjkchar+bpe";
// config.ModelConfig.BpeVocab = Path.Combine(AppCommon.AIModelFile, "sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20", "bpe.model");
// config.ModelConfig.Transducer = new OfflineTransducerModelConfig()
// {
// Decoder = Path.Combine(AppCommon.AIModelFile, "sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20", "decoder-epoch-99-avg-1.onnx"),
// Encoder = Path.Combine(AppCommon.AIModelFile, "sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20", "encoder-epoch-99-avg-1.onnx"),
// Joiner = Path.Combine(AppCommon.AIModelFile, "sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20", "joiner-epoch-99-avg-1.onnx"),
// };
//}
#endregion
#if DEBUG
config.ModelConfig.Debug = 1;
#endif
OR = new OfflineRecognizer(config);
OfflineRecognizerConfig oldConfig = new OfflineRecognizerConfig();
//采样率
oldConfig.FeatConfig.SampleRate = 16000;
oldConfig.FeatConfig.FeatureDim = 80;
oldConfig.ModelConfig.Tokens = Path.Combine(AppCommon.AIModelFile, AIModelVersion_270717, "tokens.txt");
oldConfig.ModelConfig.SenseVoice.Model = Path.Combine(AppCommon.AIModelFile, AIModelVersion_270717, "model.onnx");
oldConfig.ModelConfig.SenseVoice.UseInverseTextNormalization = 1;
//反转文本规范化规则 fst 的路径
//config.RuleFsts = Path.Combine(AppCommon.AIModelFile, "itn_subject_sx.fst");
oldConfig.ModelConfig.SenseVoice.Language = "zh";
//模型类型
oldConfig.ModelConfig.ModelType = string.Empty;
oldConfig.ModelConfig.NumThreads = numThreads;
oldConfig.ModelConfig.Provider = "cpu";
OR_old = new OfflineRecognizer(oldConfig);
VADModelConfig = new VadModelConfig();
VADModelConfig.SileroVad.Model = Path.Combine(AppCommon.AIModelFile, AIModelVersion_270717, "silero_vad.onnx");
VADModelConfig.Debug = 0;
}
/// <summary>
@ -146,137 +97,42 @@ namespace VideoAnalysisCore.AICore.SherpaOnnx
/// </summary>
/// <param name="s"></param>
/// <returns></returns>
public async Task<List<SenseVoiceRes>> RunTask(Stream s)
public List<SenseVoiceRes> RunTask(Stream s)
{
if (s is null)
throw new Exception("音频路径 is null");
return await TaskHandle(new WaveReader(s), null);
if (s is null) throw new Exception("音频路径 is null");
return serviceProvider.GetRequiredService<SherpaVad>()
.TaskHandle(new WaveReader(s), null, SoundHandle, SherpaVadVersion.silero_vad_v5);
}
/// <summary>
/// 获取语音字幕
/// </summary>
/// <param name="task"></param>
/// <returns></returns>
public async Task RunTask(string task)
public Task RunTask(string task)
{
var filePath = Path.Combine(task.LocalPath(), "task.wav");
if (string.IsNullOrEmpty(filePath) || !File.Exists(filePath))
throw new Exception("task 音频路径未找到");
await TaskHandle(new WaveReader(filePath), task);
}
serviceProvider.GetRequiredService<SherpaVad>()
.TaskHandle(new WaveReader(filePath), null, SoundHandle, SherpaVadVersion.silero_vad_v5);
/// <summary>
/// 任务处理
/// </summary>
/// <param name="reader">Wave</param>
/// <param name="task">任务id [默认Null]</param>
/// <returns></returns>
/// <exception cref="Exception"></exception>
public async Task<List<SenseVoiceRes>> TaskHandle(WaveReader reader, string? task )
{
if (OR is null)
Init();
int numSamples = reader.Samples.Length;
int windowSize = VADModelConfig.SileroVad.WindowSize;
int sampleRate = VADModelConfig.SampleRate;
int numIter = numSamples / windowSize;
var totalSecond = numSamples / (float)sampleRate;
var res = new List<SenseVoiceRes>(500);
using var VAD = new VoiceActivityDetector(VADModelConfig, bufferSizeInSeconds: 20);
for (int i = 0; i != numIter; ++i)
{
int start = i * windowSize;
float[] samples = new float[windowSize];
Array.Copy(reader.Samples, start, samples, 0, windowSize);
VAD.AcceptWaveform(samples);
//Memory<float> samples = new float[windowSize];
//Memory<float> sourceSpan = reader.Samples.AsMemory(start, windowSize);
//sourceSpan.CopyTo(samples);
//VAD.AcceptWaveform(samples.ToArray());
//是否检测到语音
if (VAD.IsSpeechDetected())
{
//获取最新的发言片段
while (!VAD.IsEmpty())
{
var p = await ReadNext(VAD,res, totalSecond);
if (p != null) redisManager.SetTaskProgress(task, p + "%");
}
}
}
VAD.Flush();
while (!VAD.IsEmpty())
{
var p = await ReadNext(VAD, res, totalSecond);
if(p!= null) redisManager.SetTaskProgress(task, p + "%");
}
//如果携带任务ID
if (!string.IsNullOrEmpty(task))
{
await redisManager.AddTaskLog(task, "==> SenseVoice 字幕数量" + res.Count);
var captionsStr = res.ToJson();
await videoTaskDB.AsUpdateable()
.SetColumns(it => it.Captions == captionsStr)
.Where(it => it.Id == long.Parse(task))
.ExecuteCommandAsync();
await redisManager.Redis.HMSetAsync(RedisExpandKey.Task(task), "Captions", res);
//分析完成视频字幕后继续接收任务
//redisManager.NewTask();
}
return res;
return Task.CompletedTask;
}
/// <summary>
/// 处理vad 下一个切片
/// 获取语音字幕
/// </summary>
/// <param name="VAD"></param>
/// <param name="res">字幕处理后写入数组</param>
/// <param name="totalSecond">总时长</param>
/// <param name="progressCallback">任务回调</param>
/// <returns></returns>
public async Task<double?> ReadNext(VoiceActivityDetector VAD, List<SenseVoiceRes> res, float totalSecond)
/// <param name="sampleRate">采样率</param>
/// <param name="samples">采样值(样品)</param>
/// <returns>结果流</returns>
public OfflineStream SoundHandle(int sampleRate, float[] samples)
{
var segment = VAD.Front();
var sampleRate = VADModelConfig.SampleRate;
var sampleRateF = (float)VADModelConfig.SampleRate;
float startTime = segment.Start / sampleRateF;
float duration = segment.Samples.Length / sampleRateF;
using var stream = OR.CreateStream();
stream.AcceptWaveform(sampleRate, segment.Samples);
var stream = OR.CreateStream();
stream.AcceptWaveform(sampleRate, samples);
OR.Decode(stream);
//old
using var stream1 = OR_old.CreateStream();
stream1.AcceptWaveform(sampleRate, segment.Samples);
OR.Decode(stream1);
if (stream.Result.Text != stream1.Result.Text)
{
Console.WriteLine("=>" + (float)Math.Round(startTime, 2, MidpointRounding.AwayFromZero));
Console.WriteLine("新=>" + stream.Result.Text);
Console.WriteLine("旧=>" + stream1.Result.Text);
}
Console.WriteLine();
double? resP =null;
if (!string.IsNullOrEmpty(stream.Result.Text))
{
var text = stream.Result.Text.Trim();
if (text.Length == 1 && text == "。")// 检查字符是否只有一个句号
{
VAD.Pop();
return resP;
}
res.Add(new()
{
Text = stream.Result.Text,
Start = (float)Math.Round(startTime, 2, MidpointRounding.AwayFromZero),
End = (float)Math.Round(startTime + duration, 2, MidpointRounding.AwayFromZero),
});
resP = Math.Round((double)(startTime + duration) / (totalSecond) * 100, 2);
}
VAD.Pop();
return resP;
return stream;
}
}
}

210
VideoAnalysisCore/AICore/SherpaOnnx/SherpaVad.cs Normal file
View File

@ -0,0 +1,210 @@
using Microsoft.Extensions.DependencyInjection;
using Microsoft.Extensions.Options;
using SherpaOnnx;
using SqlSugar;
using SqlSugar.IOC;
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.IO;
using System.Linq;
using System.Text;
using System.Text.Json;
using System.Text.RegularExpressions;
using System.Threading.Tasks;
using VideoAnalysisCore.Common;
using VideoAnalysisCore.Model;
using VideoAnalysisCore.Model.Enum;
using static System.Net.WebRequestMethods;
namespace VideoAnalysisCore.AICore.SherpaOnnx
{
public static class SherpaVadExpand
{
/// <summary>
/// 添加 Vad 语言切片
/// </summary>
/// <param name="services"></param>
public static void AddSherpaVadExpand(this IServiceCollection services)
{
services.AddTransient<SherpaVad>();
}
}
/// <summary>
/// 语音切片服务的版本
/// </summary>
public class SherpaVadVersion
{
public const string silero_vad_v4 = "silero_vad_v4.onnx";
public const string silero_vad_v5 = "silero_vad_v5.onnx";
/// <summary>
/// ten_vad (324 kb版本)
/// </summary>
public const string ten_vad_324 = "ten-vad.onnx";
}
/// <summary>
/// 语音切片服务
/// </summary>
public class SherpaVad
{
static VadModelConfig VADModelConfig = default!;
private readonly RedisManager redisManager;
private readonly IServiceProvider serviceProvider;
private readonly VoiceActivityDetector vad;
private Func<int, float[], OfflineStream> Callback;
public SherpaVad(RedisManager redisManager, IServiceProvider serviceProvider)
{
this.redisManager = redisManager;
this.serviceProvider = serviceProvider;
VADModelConfig = new VadModelConfig();
VADModelConfig.SampleRate = 16000;
VADModelConfig.NumThreads = 1;
VADModelConfig.Provider = "cpu";
#if DEBUG
VADModelConfig.Debug = 1;
#endif
VADModelConfig.SileroVad = new SileroVadModelConfig();
VADModelConfig.TenVad = new TenVadModelConfig();
}
/// <summary>
/// 初始化 SenseVoice
/// </summary>
/// <param name="func">vad识别成功后触发后回调</param>
/// <param name="vadVersion">版本采用 <see cref="SherpaVadVersion.silero_vad_v5"/> </param>
/// <param name="numThreads">默认1线程</param>
/// <param name="useGPU">是否使用gpu 报错请看安装CUDA环境<see cref="https://k2-fsa.github.io/sherpa/onnx/pretrained_models/whisper/large-v3.html#run-with-gpu-float32"/></param>
private void Init(Func<int, float[], OfflineStream> func, string vadVersion = SherpaVadVersion.silero_vad_v5, int numThreads = 1, bool useGPU = false)
{
VADModelConfig.NumThreads = numThreads;
VADModelConfig.Provider = useGPU? "cuda" : "cpu";
var path = Path.Combine(AppCommon.AIModelFile, "vad", SherpaVadVersion.silero_vad_v5);
switch (vadVersion)
{
case SherpaVadVersion.silero_vad_v4:
case SherpaVadVersion.silero_vad_v5:
VADModelConfig.SileroVad.Model = path;
break;
case SherpaVadVersion.ten_vad_324:
VADModelConfig.TenVad.Model = path;
break;
default:
break;
}
Callback = func;
}
/// <summary>
/// 任务处理
/// </summary>
/// <param name="reader">Wave</param>
/// <param name="func">vad识别成功后触发后回调</param>
/// <param name="vadVersion">版本采用 <see cref="SherpaVadVersion.silero_vad_v5"/> </param>
/// <param name="numThreads">默认1线程</param>
/// <param name="useGPU">是否使用gpu 报错请看安装CUDA环境<see cref="https://k2-fsa.github.io/sherpa/onnx/pretrained_models/whisper/large-v3.html#run-with-gpu-float32"/></param>
/// <param name="task">任务id [默认Null]</param>
/// <returns></returns>
/// <exception cref="Exception"></exception>
public List<SenseVoiceRes> TaskHandle(WaveReader reader, string? task,Func<int, float[], OfflineStream> func, string vadVersion = SherpaVadVersion.silero_vad_v5, int numThreads = 1, bool useGPU = false )
{
Init(func, vadVersion, numThreads, useGPU);
// 使用 Span 操作原始数据
ReadOnlySpan<float> allSamples = reader.Samples.AsSpan();
int numSamples = allSamples.Length;
int windowSize = VADModelConfig.SileroVad.WindowSize;
int sampleRate = VADModelConfig.SampleRate;
int numIter = numSamples / windowSize;
var totalSecond = numSamples / (float)sampleRate;
var res = new List<SenseVoiceRes>(500);
using var VAD = new VoiceActivityDetector(VADModelConfig, bufferSizeInSeconds: 30);
// 优化：复用缓冲区，避免在循环中重复分配内存
float[] buffer = new float[windowSize];
for (int i = 0; i != numIter; ++i)
{
int start = i * windowSize;
// 使用 Span 高效复制数据到固定缓冲区
allSamples.Slice(start, windowSize).CopyTo(buffer);
VAD.AcceptWaveform(buffer);
//是否检测到语音
if (VAD.IsSpeechDetected())
{
//获取最新的发言片段
while (!VAD.IsEmpty())
{
var p = ReadNext(VAD,res, totalSecond);
if (p != null) redisManager.SetTaskProgress(task, p + "%");
}
}
}
VAD.Flush();
while (!VAD.IsEmpty())
{
var p = ReadNext(VAD, res, totalSecond);
if(p!= null) redisManager.SetTaskProgress(task, p + "%");
}
//如果携带任务ID
if (!string.IsNullOrEmpty(task))
{
_ = redisManager.AddTaskLog(task, "==> SenseVoice 字幕数量" + res.Count);
var captionsStr = res.ToJson();
_ = serviceProvider.GetRequiredService<Repository<VideoTask>>()
.AsUpdateable()
.SetColumns(it => it.Captions == captionsStr)
.Where(it => it.Id == long.Parse(task))
.ExecuteCommandAsync();
_ = redisManager.Redis.HMSetAsync(RedisExpandKey.Task(task), "Captions", res);
//分析完成视频字幕后继续接收任务
//redisManager.NewTask();
}
return res;
}
/// <summary>
/// 处理vad 下一个切片
/// </summary>
/// <param name="VAD"></param>
/// <param name="res">字幕处理后写入数组</param>
/// <param name="totalSecond">总时长</param>
/// <returns></returns>
public double? ReadNext(VoiceActivityDetector VAD, List<SenseVoiceRes> res, float totalSecond)
{
var segment = VAD.Front();
var sampleRate = VADModelConfig.SampleRate;
var sampleRateF = (float)VADModelConfig.SampleRate;
float startTime = segment.Start / sampleRateF;
float duration = segment.Samples.Length / sampleRateF;
using var stream = Callback(sampleRate, segment.Samples);
double? resP =null;
if (!string.IsNullOrEmpty(stream.Result.Text))
{
var text = stream.Result.Text.Trim();
if (text.Length == 1 && text == "。")// 检查字符是否只有一个句号
{
VAD.Pop();
return resP;
}
res.Add(new()
{
Text = stream.Result.Text,
Start = (float)Math.Round(startTime, 2, MidpointRounding.AwayFromZero),
End = (float)Math.Round(startTime + duration, 2, MidpointRounding.AwayFromZero),
});
resP = Math.Round((double)(startTime + duration) / (totalSecond) * 100, 2);
}
VAD.Pop();
return resP;
}
}
}

8
VideoAnalysisCore/Controllers/VideoTaskController.cs
View File

@ -129,7 +129,7 @@ namespace VideoAnalysisCore.Controllers
using HttpClient client = new HttpClient();
// 发送GET请求获取网络文件流
using var networkStream = await client.GetStreamAsync(url);
var res = await senseVoice.RunTask(networkStream);
var res = senseVoice.RunTask(networkStream);
return Ok(res);
}
catch (Exception ex)
@ -143,11 +143,11 @@ namespace VideoAnalysisCore.Controllers
/// <param name="file">文件流</param>
/// <returns></returns>
[HttpPost(Name = "AudioRecognition")]
public async Task<IActionResult> AudioRecognition(IFormFile file)
public IActionResult AudioRecognition(IFormFile file)
{
using var s = file.OpenReadStream();
var res = await senseVoice.RunTask(s);
return Ok(res);
var res = senseVoice.RunTask(s);
return Ok(res);
}

2
VideoAnalysisCore/VideoAnalysisCore.csproj
View File

@ -71,7 +71,7 @@
<PackageReference Include="Microsoft.Extensions.DependencyModel" Version="7.0.0" />
<PackageReference Include="Microsoft.Extensions.Http" Version="8.0.0" />
<PackageReference Include="Newtonsoft.Json" Version="13.0.3" />
<PackageReference Include="org.k2fsa.sherpa.onnx" Version="1.12.20" />
<PackageReference Include="org.k2fsa.sherpa.onnx" Version="1.12.21" />
<PackageReference Include="SixLabors.ImageSharp" Version="3.1.7" />
<PackageReference Include="SqlSugar.IOC" Version="2.0.0" />
<PackageReference Include="SqlSugarCore" Version="5.1.4.205" />