From 6359c9f50dd71cc3ba39f6ae3552460d532d0f25 Mon Sep 17 00:00:00 2001
From: Cogent Apps <cogentapps@fastmail.com>
Date: Mon, 20 Mar 2023 21:03:12 +0000
Subject: [PATCH] fallback to whisper when speechrecognition is not available

---
 app/src/components/input.tsx                  | 25 +++++++++++++------
 app/src/components/settings/user.tsx          |  5 ++--
 ...types.d.ts => speech-recognition-types.ts} | 13 +++++++---
 3 files changed, 29 insertions(+), 14 deletions(-)
 rename app/src/{speech-recognition-types.d.ts => speech-recognition-types.ts} (92%)

diff --git a/app/src/components/input.tsx b/app/src/components/input.tsx
index e41dd775..1ac7dd7d 100644
--- a/app/src/components/input.tsx
+++ b/app/src/components/input.tsx
@@ -8,8 +8,8 @@ import { useAppContext } from '../context';
 import { useAppDispatch, useAppSelector } from '../store';
 import { selectMessage, setMessage } from '../store/message';
 import { selectTemperature } from '../store/parameters';
-import { openSystemPromptPanel, openTemperaturePanel } from '../store/settings-ui';
-import { speechRecognition } from '../speech-recognition-types.d'
+import { openOpenAIApiKeyPanel, openSystemPromptPanel, openTemperaturePanel } from '../store/settings-ui';
+import { speechRecognition, supportsSpeechRecognition } from '../speech-recognition-types'
 import MicRecorder from 'mic-recorder-to-mp3';
 import { selectUseOpenAIWhisper, selectOpenAIApiKey } from '../store/api-keys';
 import { Mp3Encoder } from 'lamejs';
@@ -109,7 +109,7 @@ export default function MessageInput(props: MessageInputProps) {
         console.error('speech recognition error', e);
 
         try {
-            speechRecognition.stop();
+            speechRecognition?.stop();
         } catch (e) {
         }
 
@@ -122,14 +122,19 @@ export default function MessageInput(props: MessageInputProps) {
     }, [recorder]);
 
     const onSpeechStart = useCallback(() => {
+        if (!openAIApiKey) {
+            dispatch(openOpenAIApiKeyPanel());
+            return false;
+        }
+
         try {
             if (!recording) {
                 setRecording(true);
 
                 // if we are using whisper, the we will just record with the browser and send the api when done 
-                if (useOpenAIWhisper) {
+                if (useOpenAIWhisper || !supportsSpeechRecognition) {
                     recorder.start().catch(onSpeechError);
-                } else {
+                } else if (speechRecognition) {
                     const initialMessage = message;
 
                     speechRecognition.continuous = true;
@@ -146,10 +151,12 @@ export default function MessageInput(props: MessageInputProps) {
                     };
 
                     speechRecognition.start();
+                } else {
+                    onSpeechError(new Error('not supported'));
                 }
             } else {
                 setRecording(false);
-                if (useOpenAIWhisper) {
+                if (useOpenAIWhisper || !supportsSpeechRecognition) {
                     setTranscribing(true);
                     const mp3 = recorder.stop().getMp3();
 
@@ -185,14 +192,16 @@ export default function MessageInput(props: MessageInputProps) {
                         }
 
                     }).catch(onSpeechError);
-                } else {
+                } else if (speechRecognition) {
                     speechRecognition.stop();
+                } else {
+                    onSpeechError(new Error('not supported'));
                 }
             }
         } catch (e) {
             onSpeechError(e);
         }
-    }, [recording, message, dispatch]);
+    }, [recording, message, dispatch, onSpeechError, openAIApiKey]);
 
 
     const onKeyDown = useCallback((e: React.KeyboardEvent<HTMLTextAreaElement>) => {
diff --git a/app/src/components/settings/user.tsx b/app/src/components/settings/user.tsx
index 6b6d1303..474e3be1 100644
--- a/app/src/components/settings/user.tsx
+++ b/app/src/components/settings/user.tsx
@@ -6,6 +6,7 @@ import { useAppDispatch, useAppSelector } from "../../store";
 import { selectOpenAIApiKey, setOpenAIApiKeyFromEvent, selectUseOpenAIWhisper, setUseOpenAIWhisperFromEvent } from "../../store/api-keys";
 import { selectSettingsOption } from "../../store/settings-ui";
 import { FormattedMessage, useIntl } from "react-intl";
+import { supportsSpeechRecognition } from "../../speech-recognition-types";
 
 export default function UserOptionsTab(props: any) {
     const option = useAppSelector(selectSettingsOption);
@@ -31,11 +32,11 @@ export default function UserOptionsTab(props: any) {
                     </a>
                 </p>
 
-                <Checkbox
+                {supportsSpeechRecognition && <Checkbox
                     style={{ marginTop: '1rem' }}
                     id="use-openai-whisper-api" checked={useOpenAIWhisper!} onChange={onUseOpenAIWhisperChange}
                     label="Use the OpenAI Whisper API for speech recognition."
-                />
+                />}
 
                 <p>
                     <FormattedMessage defaultMessage="Your API key is stored only on this device and never transmitted to anyone except OpenAI." />
diff --git a/app/src/speech-recognition-types.d.ts b/app/src/speech-recognition-types.ts
similarity index 92%
rename from app/src/speech-recognition-types.d.ts
rename to app/src/speech-recognition-types.ts
index 8e205a5b..89eee651 100644
--- a/app/src/speech-recognition-types.d.ts
+++ b/app/src/speech-recognition-types.ts
@@ -122,12 +122,17 @@ declare global {
     }
 }
 
-let speechRecognition: SpeechRecognition
+let speechRecognition: SpeechRecognition | null = null;
 
 if (window.SpeechRecognition) {
     speechRecognition = new SpeechRecognition()
-} else {
-    speechRecognition = new webkitSpeechRecognition()
+} else if ((window as any).webkitSpeechRecognition) {
+    speechRecognition = new (window as any).webkitSpeechRecognition() as SpeechRecognition;
 }
 
-export { speechRecognition }
\ No newline at end of file
+const supportsSpeechRecognition = speechRecognition !== null;
+
+export {
+    speechRecognition,
+    supportsSpeechRecognition,
+}
\ No newline at end of file