Fix DiffSinger rendering (Phonemizer & Multi-Speaker) & Misc.

OpenUtau.Core/DiffSinger/DiffSingerBasePhonemizer.cs

@@ -1,4 +1,4 @@
-using System;
+using System;
 using System.Collections.Generic;
 using System.IO;
 using System.Linq;
@@ -49,10 +49,11 @@ private bool _executeSetSinger(USinger singer) {
             if (singer == null) {
                 throw new Exception("Singer is null.");
             }
-            if(singer.Location == null){
-                throw new Exception("Singer location is null.");
+            if (File.Exists(Path.Join(singer.Location, "dsdur", "dsconfig.yaml"))) {
+                rootPath = Path.Combine(singer.Location, "dsdur");
+            } else {
+                rootPath = singer.Location;
             }
-            rootPath = Path.Combine(singer.Location, "dsdur");
             //Load Config
             var configPath = Path.Join(rootPath, "dsconfig.yaml");
             try {
@@ -207,9 +208,10 @@ string GetSpeakerAtIndex(Note note, int index) {
                 speaker = singer.Subbanks.FirstOrDefault();
             }
             if (speaker is null) {
-                throw new Exception(
-                    $"No subbanks defined for singer \"{singer.Name}\". " +
-                    "Please check the singer's configuration.");
+                if (dsConfig.speakers != null && dsConfig.speakers.Count > 0) {
+                    return dsConfig.speakers[0];
+                }
+                return "";
             }
             return speaker.Suffix;
         }

OpenUtau.Core/DiffSinger/DiffSingerSpeakerEmbedManager.cs

@@ -22,6 +22,7 @@ public DiffSingerSpeakerEmbedManager(DsConfig dsConfig, string rootPath) {
             this.dsConfig = dsConfig;
             this.rootPath = rootPath;
         }
+
         public NDArray loadSpeakerEmbed(string speaker) {
             string path = Path.Join(rootPath, speaker + ".emb");
             if(File.Exists(path)) {
@@ -50,14 +51,16 @@ public NDArray getSpeakerEmbeds() {
 
         public bool IsVoiceColorCurve(string abbr, out int subBankId) {
             subBankId = 0;
-            if (abbr.StartsWith(VoiceColorHeader) && int.TryParse(abbr.Substring(2), out subBankId)) {;
+            if (abbr.StartsWith(VoiceColorHeader) && int.TryParse(abbr.Substring(2), out subBankId)) {
                 subBankId -= 1;
                 return true;
             } else {
                 return false;
             }
         }
 
+        static readonly HashSet<string> warnedMissingSpeakerSuffixes = new();
+
         public int getSpeakerIndexBySuffix(string suffix) {
             var speakerIndex = dsConfig.speakers.IndexOf(suffix);
             if (speakerIndex >= 0) {
@@ -72,69 +75,124 @@ public int getSpeakerIndexBySuffix(string suffix) {
             if (speakerIndex >= 0) {
                 return speakerIndex;
             }
-            if (dsConfig.speakers.Count == 0) {
+            if (dsConfig.speakers == null || dsConfig.speakers.Count == 0) {
                 throw new InvalidOperationException(
                     "Subbanks are defined in character.yaml but \"speakers\" is empty in dsconfig.yaml.");
             }
-            Log.Warning(
-                $"Speaker suffix \"{suffix}\" not found in dsConfig.speakers, falling back to first speaker. " +
-                $"Candidates: {string.Join(',', dsConfig.speakers)}.");
+            var fallback = dsConfig.speakers[0];
+            var warnKey = $"{rootPath}|{suffix}|{fallback}";
+            lock (warnedMissingSpeakerSuffixes) {
+                if (warnedMissingSpeakerSuffixes.Add(warnKey)) {
+                    Log.Warning(
+                        "Speaker suffix \"{Suffix}\" not found in dsConfig.speakers ({Candidates}). Falling back to \"{Fallback}\".",
+                        suffix,
+                        string.Join(", ", dsConfig.speakers),
+                        fallback);
+                }
+            }
             return 0;
         }
 
         //used by phonemizer (duration model)
-        public Tensor<float> PhraseSpeakerEmbedByPhone(string[] speakerByPhone){
+        public Tensor<float> PhraseSpeakerEmbedByPhone(string[] speakerByPhone) {
             var hiddenSize = dsConfig.hiddenSize;
             var speakerEmbeds = getSpeakerEmbeds();
+            if (speakerEmbeds == null) {
+                return null;
+            }
+            int speakerCount = dsConfig.speakers.Count;
+            var speakerEmbedArrays = new float[speakerCount][];
+            for (int spk = 0; spk < speakerCount; spk++) {
+                speakerEmbedArrays[spk] = speakerEmbeds[":", spk].ToArray<float>();
+            }
+
             var totalPhones = speakerByPhone.Length;
-            NDArray spkCurves = np.zeros<float>(totalPhones, dsConfig.speakers.Count);
-            foreach(int phoneId in Enumerable.Range(0,totalPhones)) {
+            var result = new float[totalPhones * hiddenSize];
+
+            for (int phoneId = 0; phoneId < totalPhones; phoneId++) {
                 var spkId = getSpeakerIndexBySuffix(speakerByPhone[phoneId]);
-                spkCurves[phoneId, spkId] = 1;
+                var embed = speakerEmbedArrays[spkId];
+                var dest = result.AsSpan(phoneId * hiddenSize, hiddenSize);
+                embed.AsSpan().CopyTo(dest);
             }
-            var spkEmbedResult = np.dot(spkCurves, speakerEmbeds.T);
-            var spkEmbedTensor = new DenseTensor<float>(spkEmbedResult.ToArray<float>(), 
-                new int[] { totalPhones, hiddenSize })
+
+            return new DenseTensor<float>(result, new int[] { totalPhones, hiddenSize })
                 .Reshape(new int[] { 1, totalPhones, hiddenSize });
-            return spkEmbedTensor;
         }
 
         //used by variance, pitch and acoustic
-        public Tensor<float> PhraseSpeakerEmbedByFrame(RenderPhrase phrase, IList<int> durations, float frameMs, int totalFrames, int headFrames, int tailFrames){
+        public Tensor<float> PhraseSpeakerEmbedByFrame(RenderPhrase phrase, IList<int> durations, float frameMs, int totalFrames, int headFrames, int tailFrames) {
             var singer = phrase.singer;
             var hiddenSize = dsConfig.hiddenSize;
             var speakerEmbeds = getSpeakerEmbeds();
-            //get default speaker for each phoneme
+            if (speakerEmbeds == null) {
+                return null;
+            }
+            // Per-frame CLR / phoneme suffix is always weight 1.0 ("100%").
+            // Voice-color curves add on top; then weights are normalized to a convex mix.
+            // Example: CLR=A and cl_B=100% → A:B = 1:1 (not pure B).
             var headDefaultSpk = getSpeakerIndexBySuffix(phrase.phones[0].suffix);
             var tailDefaultSpk = getSpeakerIndexBySuffix(phrase.phones[^1].suffix);
             var defaultSpkByFrame = Enumerable.Repeat(headDefaultSpk, headFrames).ToList();
             defaultSpkByFrame.AddRange(Enumerable.Range(0, phrase.phones.Length)
-                .SelectMany(phIndex => Enumerable.Repeat(getSpeakerIndexBySuffix(phrase.phones[phIndex].suffix), durations[phIndex+1])));
+                .SelectMany(phIndex => Enumerable.Repeat(
+                    getSpeakerIndexBySuffix(phrase.phones[phIndex].suffix),
+                    durations[phIndex + 1])));
             defaultSpkByFrame.AddRange(Enumerable.Repeat(tailDefaultSpk, tailFrames));
             //get speaker curves
             NDArray spkCurves = np.zeros<float>(totalFrames, dsConfig.speakers.Count);
-            foreach(var curve in phrase.curves) {
-                if(IsVoiceColorCurve(curve.Item1,out int subBankId) && subBankId < singer.Subbanks.Count) {
+            foreach (var curve in phrase.curves) {
+                if (IsVoiceColorCurve(curve.Item1, out int subBankId) && subBankId < singer.Subbanks.Count) {
                     var spkId = getSpeakerIndexBySuffix(singer.Subbanks[subBankId].Suffix);
                     spkCurves[":", spkId] += DiffSingerUtils.SampleCurve(phrase, curve.Item2, 0, 
                         frameMs, totalFrames, headFrames, tailFrames, x => x * 0.01f)
                         .Select(f => (float)f).ToArray();
                 }
             }
-            foreach(int frameId in Enumerable.Range(0,totalFrames)) {
-                //standarization
-                var spkSum = spkCurves[frameId, ":"].ToArray<float>().Sum();
-                if (spkSum > 1) {
-                    spkCurves[frameId, ":"] /= spkSum;
-                } else {
-                    spkCurves[frameId, defaultSpkByFrame[frameId]] += 1 - spkSum;
+
+            int speakerCount = dsConfig.speakers.Count;
+            var speakerEmbedArrays = new float[speakerCount][];
+            for (int spk = 0; spk < speakerCount; spk++) {
+                speakerEmbedArrays[spk] = speakerEmbeds[":", spk].ToArray<float>();
+            }
+
+            var result = new float[totalFrames * hiddenSize];
+            var weights = new float[speakerCount];
+            for (int frameId = 0; frameId < totalFrames; frameId++) {
+                Array.Clear(weights, 0, speakerCount);
+                int clrSpkId = defaultSpkByFrame[frameId];
+                weights[clrSpkId] = 1f;
+                for (int spk = 0; spk < speakerCount; spk++) {
+                    weights[spk] += (float)spkCurves[frameId, spk];
+                }
+
+                float weightSum = 0f;
+                for (int spk = 0; spk < speakerCount; spk++) {
+                    if (weights[spk] < 0f) {
+                        weights[spk] = 0f;
+                    }
+                    weightSum += weights[spk];
+                }
+                if (weightSum < 1e-8f) {
+                    weights[clrSpkId] = 1f;
+                    weightSum = 1f;
+                }
+
+                var dest = result.AsSpan(frameId * hiddenSize, hiddenSize);
+                dest.Clear();
+                for (int spk = 0; spk < speakerCount; spk++) {
+                    float w = weights[spk] / weightSum;
+                    if (w < 1e-8f) {
+                        continue;
+                    }
+                    var embed = speakerEmbedArrays[spk];
+                    for (int j = 0; j < dest.Length; j++) {
+                        dest[j] += w * embed[j];
+                    }
                 }
             }
-            var spkEmbedResult = np.dot(spkCurves, speakerEmbeds.T);
-            var spkEmbedTensor = new DenseTensor<float>(spkEmbedResult.ToArray<float>(), 
-                new int[] { totalFrames, hiddenSize })
+            return new DenseTensor<float>(result, new int[] { totalFrames, hiddenSize })
                 .Reshape(new int[] { 1, totalFrames, hiddenSize });
-            return spkEmbedTensor;
         }
     }
 }

OpenUtau.Core/DiffSinger/Phonemizers/DiffSingerG2pPhonemizer.cs

@@ -1,4 +1,4 @@
-using Serilog;
+using Serilog;
 using System;
 using System.Collections.Generic;
 using System.IO;

OpenUtauMobile/Helpers/LocalizationManager.cs

@@ -1,4 +1,4 @@
-using System;
+using System;
 using System.Collections.Generic;
 using System.Globalization;
 using System.IO;
@@ -21,7 +21,7 @@ public static class LocalizationManager
     public const string FollowSystemLanguageCode = "system";
 
     private const string ExplicitLanguageFallbackCode = "en";
-    private const string SystemLanguageFallbackCode = "zh-Hans";
+    private const string SystemLanguageFallbackCode = "en";
 
     /// <summary>Supported languages (language code => display name).</summary>
     public static readonly IReadOnlyList<(string Code, string DisplayName)> AvailableLanguages =
@@ -77,7 +77,7 @@ public static void LoadLanguage(string langCode)
 
     /// <summary>
     /// Resolves language preference to a supported resource language code.
-    /// Empty or "system" preference uses system UI language and falls back to zh-Hans.
+    /// Empty or "system" preference uses system UI language and falls back to en.
     /// Explicit codes fall back to en when unsupported.
     /// </summary>
     public static string ResolveLanguagePreference(string? preferenceCode)
@@ -93,7 +93,7 @@ public static string ResolveLanguagePreference(string? preferenceCode)
 
     /// <summary>
     /// Resolves current system language to a supported resource language code.
-    /// Falls back to zh-Hans when system language is not in the supported list.
+    /// Falls back to en when system language is not in the supported list.
     /// </summary>
     public static string ResolveSystemLanguage()
     {

OpenUtauMobile/ViewModels/ClassicSingerSetupViewModel.cs

@@ -1,4 +1,4 @@
-using System;
+using System;
 using System.Collections.ObjectModel;
 using System.IO;
 using System.Linq;
@@ -99,6 +99,22 @@ public ClassicSingerSetupViewModel(MainViewModel navigator) : base(navigator)
                         VoicebankConfig? config = LoadCharacterYaml(ArchiveFilePath);
                         MissingInfo = config == null || string.IsNullOrEmpty(config.SingerType);
 
+                        if (!MissingInfo)
+                        {
+                            // character.yaml has a SingerType — use it as default selection.
+                            var declaredType = config!.SingerType;
+                            if (SingerTypes.Contains(declaredType))
+                            {
+                                SingerType = declaredType;
+                            }
+                        }
+                        else
+                        {
+                            // No SingerType in yaml (or no yaml at all).
+                            // Mirror VoicebankLoader.LoadInfo heuristic: scan archive for config files.
+                            SingerType = DetectSingerTypeFromArchive(ArchiveFilePath);
+                        }
+
                         if (!string.IsNullOrEmpty(config?.TextFileEncoding))
                         {
                             try
@@ -291,6 +307,45 @@ private bool IsEncrypted(string archiveFilePath)
         }
     }
 
+    /// <summary>
+    /// Scans archive entries for dsconfig.yaml, enuconfig.yaml, or info.toml to infer singer type.
+    /// Mirrors the legacy detection heuristic in VoicebankLoader.LoadInfo and NeutrinoConfig.Load.
+    /// </summary>
+    private string DetectSingerTypeFromArchive(string archiveFilePath)
+    {
+        try
+        {
+            using (IArchive archive = ArchiveFactory.OpenArchive(archiveFilePath))
+            {
+                bool hasDsconfig = archive.Entries.Any(e =>
+                    Path.GetFileName(e.Key) == "dsconfig.yaml");
+                if (hasDsconfig)
+                {
+                    return "diffsinger";
+                }
+
+                bool hasEnuconfig = archive.Entries.Any(e =>
+                    Path.GetFileName(e.Key) == "enuconfig.yaml");
+                if (hasEnuconfig)
+                {
+                    return "enunu";
+                }
+
+                bool hasNeutrino = archive.Entries.Any(e =>
+                    Path.GetFileName(e.Key) == "info.toml");
+                if (hasNeutrino)
+                {
+                    return "neutrino";
+                }
+            }
+        }
+        catch
+        {
+            // Fall through to default
+        }
+        return "utau";
+    }
+
     private async Task InstallAsync()
     {
         string archiveFilePath = ArchiveFilePath;