Text to voice feature

submodules/AccountContext/Sources/MediaManager.swift

@@ -107,6 +107,9 @@ public func peerMessageMediaPlayerType(_ message: EngineMessage) -> MediaManager
                 break
             }
         }
+        if let attribute = message.attributes.first(where: { $0 is TextTranscriptionMessageAttribute }) as? TextTranscriptionMessageAttribute {
+            file = attribute.file
+        }
         return file
     }
 

submodules/AttachmentUI/Sources/AttachmentPanel.swift

@@ -1247,6 +1247,7 @@ final class AttachmentPanel: ASDisplayNode, ASScrollViewDelegate {
         }, updateHistoryFilter: { _ in
         }, updateDisplayHistoryFilterAsList: { _ in
         }, requestLayout: { _ in
+        }, startTranscribingText: { _ in
         }, chatController: {
             return nil
         }, statuses: nil)

submodules/ChatPresentationInterfaceState/Sources/ChatPanelInterfaceInteraction.swift

@@ -175,6 +175,7 @@ public final class ChatPanelInterfaceInteraction {
     public let openBoostToUnrestrict: () -> Void
     public let updateVideoTrimRange: (Double, Double, Bool, Bool) -> Void
     public let requestLayout: (ContainedViewLayoutTransition) -> Void
+    public let startTranscribingText: (Message) -> Void
     public let chatController: () -> ViewController?
     public let statuses: ChatPanelInterfaceInteractionStatuses?
 
@@ -292,6 +293,7 @@ public final class ChatPanelInterfaceInteraction {
         updateHistoryFilter: @escaping ((ChatPresentationInterfaceState.HistoryFilter?) -> ChatPresentationInterfaceState.HistoryFilter?) -> Void,
         updateDisplayHistoryFilterAsList: @escaping (Bool) -> Void,
         requestLayout: @escaping (ContainedViewLayoutTransition) -> Void,
+        startTranscribingText: @escaping (Message) -> Void,
         chatController: @escaping () -> ViewController?,
         statuses: ChatPanelInterfaceInteractionStatuses?
     ) {
@@ -408,7 +410,7 @@ public final class ChatPanelInterfaceInteraction {
         self.updateHistoryFilter = updateHistoryFilter
         self.updateDisplayHistoryFilterAsList = updateDisplayHistoryFilterAsList
         self.requestLayout = requestLayout
-
+        self.startTranscribingText = startTranscribingText
         self.chatController = chatController
         self.statuses = statuses
     }
@@ -532,6 +534,7 @@ public final class ChatPanelInterfaceInteraction {
         }, updateHistoryFilter: { _ in
         }, updateDisplayHistoryFilterAsList: { _ in
         }, requestLayout: { _ in
+        }, startTranscribingText: { _ in
         }, chatController: {
             return nil
         }, statuses: nil)

submodules/GalleryData/Sources/GalleryData.swift

@@ -175,6 +175,10 @@ public func chatMessageGalleryControllerData(context: AccountContext, chatLocati
         }
     }
 
+    if let attribute = message.attributes.first(where: { $0 is TextTranscriptionMessageAttribute }) as? TextTranscriptionMessageAttribute {
+        galleryMedia = attribute.file
+    }
+
     var stream = false
     var autoplayingVideo = false
     var landscape = false

submodules/TelegramCore/Sources/Account/AccountManager.swift

@@ -208,6 +208,7 @@ private var declaredEncodables: Void = {
     declareEncodable(SendAsMessageAttribute.self, f: { SendAsMessageAttribute(decoder: $0) })
     declareEncodable(ForwardVideoTimestampAttribute.self, f: { ForwardVideoTimestampAttribute(decoder: $0) })
     declareEncodable(AudioTranscriptionMessageAttribute.self, f: { AudioTranscriptionMessageAttribute(decoder: $0) })
+    declareEncodable(TextTranscriptionMessageAttribute.self, f: { TextTranscriptionMessageAttribute(decoder: $0) })
     declareEncodable(NonPremiumMessageAttribute.self, f: { NonPremiumMessageAttribute(decoder: $0) })
     declareEncodable(TelegramExtendedMedia.self, f: { TelegramExtendedMedia(decoder: $0) })
     declareEncodable(TelegramPeerUsername.self, f: { TelegramPeerUsername(decoder: $0) })

submodules/TelegramCore/Sources/State/AccountStateManagementUtils.swift

@@ -3926,6 +3926,9 @@ func replayFinalState(
                                 updatedAttributes.append(translation)
                             }
                         }
+                        if let transcription = previousMessage.attributes.first(where: { $0 is TextTranscriptionMessageAttribute }) as? TextTranscriptionMessageAttribute {
+                            updatedAttributes.append(transcription)
+                        }
                     }
 
                     if let previousFactCheckAttribute = previousMessage.attributes.first(where: { $0 is FactCheckMessageAttribute }) as? FactCheckMessageAttribute, let updatedFactCheckAttribute = message.attributes.first(where: { $0 is FactCheckMessageAttribute }) as? FactCheckMessageAttribute {

submodules/TelegramCore/Sources/SyncCore/SyncCore_TextTranscriptionMessageAttribute.swift

@@ -0,0 +1,61 @@
+//
+//  SyncCore_TextTranscriptionMessageAttribute.swift
+//  Telegram
+//
+//  Created by Dmitry Bolonikov on 7.04.25.
+//
+
+import Postbox
+
+public class TextTranscriptionMessageAttribute: MessageAttribute, Equatable {
+    public let id: Int64
+    public let visible: Bool
+    public let downloading: Bool
+    public let file: TelegramMediaFile
+
+    public var associatedPeerIds: [PeerId] {
+        return []
+    }
+
+    public init(
+        id: Int64,
+        visible: Bool,
+        downloading: Bool,
+        file: TelegramMediaFile
+    ) {
+        self.id = id
+        self.visible = visible
+        self.downloading = downloading
+        self.file = file
+    }
+
+    required public init(decoder: PostboxDecoder) {
+        self.id = decoder.decodeInt64ForKey("id", orElse: 0)
+        self.visible = decoder.decodeBoolForKey("visible", orElse: false)
+        self.downloading = decoder.decodeBoolForKey("downloading", orElse: false)
+        self.file = decoder.decodeObjectForKey("file") as! TelegramMediaFile
+    }
+
+    public func encode(_ encoder: PostboxEncoder) {
+        encoder.encodeInt64(self.id, forKey: "id")
+        encoder.encodeBool(self.visible, forKey: "visible")
+        encoder.encodeBool(self.downloading, forKey: "downloading")
+        encoder.encodeObject(file, forKey: "file")
+    }
+
+    public static func ==(lhs: TextTranscriptionMessageAttribute, rhs: TextTranscriptionMessageAttribute) -> Bool {
+        if lhs.id != rhs.id {
+            return false
+        }
+        if lhs.visible != rhs.visible {
+            return false
+        }
+        if lhs.file != rhs.file {
+            return false
+        }
+        if lhs.downloading != rhs.downloading {
+            return false
+        }
+        return true
+    }
+}

submodules/TelegramCore/Sources/TelegramEngine/Messages/TelegramEngineMessages.swift

@@ -581,6 +581,10 @@ public extension TelegramEngine {
             |> ignoreValues
         }
 
+        public func transcribeText(messageId: MessageId) -> Signal<EngineTextTranscriptionResult, NoError> {
+            _internal_transcribeText(postbox: self.account.postbox, network: self.account.network, messageId: messageId)
+        }
+
         public func storeLocallyDerivedData(messageId: MessageId, data: [String: CodableEntry]) -> Signal<Never, NoError> {
             return self.account.postbox.transaction { transaction -> Void in
                 transaction.updateMessage(messageId, update: { currentMessage in

submodules/TelegramCore/Sources/TelegramEngine/Messages/TextTranscription.swift

@@ -0,0 +1,156 @@
+//
+//  TextTranscription.swift
+//  Telegram
+//
+//  Created by Dmitry Bolonikov on 7.04.25.
+//
+
+import Foundation
+import Postbox
+import SwiftSignalKit
+import TelegramApi
+import MtProtoKit
+
+public enum EngineTextTranscriptionResult {
+    case transcribing
+    case finished
+}
+
+private enum InternalTextTranscriptionResult {
+    case alreadyTranscribed(TextTranscriptionMessageAttribute)
+    case startTranscribing(TextTranscriptionMessageAttribute)
+    case transcribed(TextTranscriptionMessageAttribute)
+    case error
+}
+
+func _internal_transcribeText(postbox: Postbox, network: Network, messageId: MessageId) -> Signal<EngineTextTranscriptionResult, NoError> {
+    return postbox.transaction { transaction -> Message? in
+        transaction.getMessage(messageId)
+    }
+    |> mapToSignal { message -> Signal<InternalTextTranscriptionResult, NoError> in
+        guard let message else {
+            return .single(.error)
+        }
+
+        if let attribute = message.attributes.first(where: { $0 is TextTranscriptionMessageAttribute }) as? TextTranscriptionMessageAttribute {
+            return .single(.alreadyTranscribed(attribute))
+        }
+
+        return Signal { subscriber in
+
+            let fileId = Int64.random(in: Int64.min...Int64.max)
+            let resource = LocalFileMediaResource(fileId: fileId)
+
+            let mediaId = MediaId(namespace: Namespaces.Media.LocalFile, id: Int64.random(in: Int64.min...Int64.max))
+
+            let voiceAttributes: [TelegramMediaFileAttribute] = [.Audio(isVoice: true, duration: 0, title: nil, performer: nil, waveform: nil)]
+
+            let file = TelegramMediaFile(
+                fileId: mediaId,
+                partialReference: nil,
+                resource: resource,
+                previewRepresentations: [],
+                videoThumbnails: [],
+                immediateThumbnailData: nil,
+                mimeType: "audio/ogg",
+                size: 1,
+                attributes: voiceAttributes,
+                alternativeRepresentations: [])
+
+            let attributeId = Int64.random(in: Int64.min...Int64.max)
+            let attribute = TextTranscriptionMessageAttribute(id: attributeId,
+                                                              visible: true,
+                                                              downloading: true,
+                                                              file: file)
+
+            subscriber.putNext(.startTranscribing(attribute))
+
+            DispatchQueue.global(qos: .userInitiated).asyncAfter(deadline: .now() + 5) {
+                guard let fileUrl = Bundle.main.url(forResource: "TextToVoiceFeature", withExtension: "ogg"),
+                      let data = try? Data(contentsOf: fileUrl) else {
+                    subscriber.putNext(.error)
+                    subscriber.putCompletion()
+                    return
+                }
+
+                postbox.mediaBox.storeResourceData(resource.id, data: data)
+
+                // TODO: Fetch duration, waveform from response
+                let waveformBase64 = "DAAOAAkACQAGAAwADwAMABAADQAPABsAGAALAA0AGAAfABoAHgATABgAGQAYABQADAAVABEAHwANAA0ACQAWABkACQAOAAwACQAfAAAAGQAVAAAAEwATAAAACAAfAAAAHAAAABwAHwAAABcAGQAAABQADgAAABQAHwAAAB8AHwAAAAwADwAAAB8AEwAAABoAFwAAAB8AFAAAAAAAHwAAAAAAHgAAAAAAHwAAAAAAHwAAAAAAHwAAAAAAHwAAAAAAHwAAAAAAAAA="
+
+                let voiceAttributes: [TelegramMediaFileAttribute] = [.Audio(isVoice: true, duration: 0, title: nil, performer: nil, waveform: Data(base64Encoded: waveformBase64)!)]
+
+                let file = TelegramMediaFile(
+                    fileId: mediaId,
+                    partialReference: nil,
+                    resource: resource,
+                    previewRepresentations: [],
+                    videoThumbnails: [],
+                    immediateThumbnailData: nil,
+                    mimeType: "audio/ogg",
+                    size: Int64(data.count),
+                    attributes: voiceAttributes,
+                    alternativeRepresentations: [])
+
+                let attributeId = Int64.random(in: Int64.min...Int64.max)
+                let attribute = TextTranscriptionMessageAttribute(id: attributeId,
+                                                                  visible: true,
+                                                                  downloading: false,
+                                                                  file: file)
+
+                subscriber.putNext(.transcribed(attribute))
+                subscriber.putCompletion()
+            }
+
+            return EmptyDisposable
+        }
+    }
+    |> mapToSignal { result -> Signal<EngineTextTranscriptionResult, NoError> in
+        return postbox.transaction { transaction -> EngineTextTranscriptionResult in
+            transaction.updateMessage(messageId, update: { currentMessage in
+                var attributes = currentMessage.attributes.filter { !($0 is TextTranscriptionMessageAttribute) }
+
+                switch result {
+                case .transcribed(let attribute):
+                    attributes.append(attribute)
+
+                case .startTranscribing(let attribute):
+                    attributes.append(attribute)
+                case .alreadyTranscribed(let attribute):
+                    let updatedAttribute = TextTranscriptionMessageAttribute(id: attribute.id, visible: true, downloading: attribute.downloading, file: attribute.file)
+                    guard updatedAttribute != attribute else {
+                        return .skip
+                    }
+                    attributes.append(updatedAttribute)
+                default:
+                    return .skip
+                }
+
+                let storeForwardInfo = currentMessage.forwardInfo.flatMap(StoreMessageForwardInfo.init)
+
+                return .update(StoreMessage(
+                    id: currentMessage.id,
+                    globallyUniqueId: currentMessage.globallyUniqueId,
+                    groupingKey: currentMessage.groupingKey,
+                    threadId: currentMessage.threadId,
+                    timestamp: currentMessage.timestamp,
+                    flags: StoreMessageFlags(currentMessage.flags),
+                    tags: currentMessage.tags,
+                    globalTags: currentMessage.globalTags,
+                    localTags: currentMessage.localTags,
+                    forwardInfo: storeForwardInfo,
+                    authorId: currentMessage.author?.id,
+                    text: currentMessage.text,
+                    attributes: attributes,
+                    media: currentMessage.media))
+            })
+
+            switch result {
+            case .startTranscribing:
+                return .transcribing
+            default:
+                return .finished
+            }
+        }
+    }
+}

submodules/TelegramUI/Components/Chat/ChatMessageInteractiveFileNode/Sources/ChatMessageInteractiveFileNode.swift

@@ -65,6 +65,9 @@ public final class ChatMessageInteractiveFileNode: ASDisplayNode {
         public let layoutConstants: ChatMessageItemLayoutConstants
         public let constrainedSize: CGSize
         public let controllerInteraction: ChatControllerInteraction
+        public let alwaysDisplayTranscriptionButton: Bool
+        public let transcriptionState: AudioTranscriptionButtonComponent.TranscriptionState?
+        public let transcriptionButtonTapped: (() -> Void)?
 
         public init(
             context: AccountContext,
@@ -88,7 +91,10 @@ public final class ChatMessageInteractiveFileNode: ASDisplayNode {
             isAttachedContentBlock: Bool,
             layoutConstants: ChatMessageItemLayoutConstants,
             constrainedSize: CGSize,
-            controllerInteraction: ChatControllerInteraction
+            controllerInteraction: ChatControllerInteraction,
+            alwaysDisplayTranscriptionButton: Bool = false,
+            transcriptionState: AudioTranscriptionButtonComponent.TranscriptionState? = nil,
+            transcriptionButtonTapped: (() -> Void)? = nil
         ) {
             self.context = context
             self.presentationData = presentationData
@@ -112,6 +118,9 @@ public final class ChatMessageInteractiveFileNode: ASDisplayNode {
             self.layoutConstants = layoutConstants
             self.constrainedSize = constrainedSize
             self.controllerInteraction = controllerInteraction
+            self.alwaysDisplayTranscriptionButton = alwaysDisplayTranscriptionButton
+            self.transcriptionState = transcriptionState
+            self.transcriptionButtonTapped = transcriptionButtonTapped
         }
     }
 
@@ -789,6 +798,10 @@ public final class ChatMessageInteractiveFileNode: ASDisplayNode {
                     }
                 }
 
+                if arguments.alwaysDisplayTranscriptionButton {
+                    displayTranscribe = true
+                }
+
                 let transcribedText = forcedAudioTranscriptionText ?? transcribedText(message: arguments.message)
 
                 switch audioTranscriptionState {
@@ -805,7 +818,7 @@ public final class ChatMessageInteractiveFileNode: ASDisplayNode {
                     updatedAudioTranscriptionState = .locked
                 }
 
-                let effectiveAudioTranscriptionState = updatedAudioTranscriptionState ?? audioTranscriptionState
+                let effectiveAudioTranscriptionState = updatedAudioTranscriptionState ?? arguments.transcriptionState ?? audioTranscriptionState
 
                 var displayTrailingAnimatedDots = false
 
@@ -1356,7 +1369,11 @@ public final class ChatMessageInteractiveFileNode: ASDisplayNode {
                                                 guard let strongSelf = self else {
                                                     return
                                                 }
-                                                strongSelf.transcribe()
+                                                if let action = arguments.transcriptionButtonTapped {
+                                                    action()
+                                                } else {
+                                                    strongSelf.transcribe()
+                                                }
                                             }
                                         )),
                                         environment: {},

submodules/TelegramUI/Components/Chat/ChatMessageTextBubbleContentNode/BUILD

@@ -35,6 +35,7 @@ swift_library(
         "//submodules/TelegramUI/Components/Chat/ShimmeringLinkNode",
         "//submodules/TelegramUI/Components/Chat/ChatMessageItemCommon",
         "//submodules/TelegramUI/Components/Chat/MessageQuoteComponent",
+        "//submodules/TelegramUI/Components/Chat/ChatMessageInteractiveFileNode",
         "//submodules/TelegramUI/Components/TextLoadingEffect",
         "//submodules/TelegramUI/Components/ChatControllerInteraction",
         "//submodules/TelegramUI/Components/InteractiveTextComponent",