Add IBM as speech provider (#160)

* add ibm speech

* remove multimedia from recognizer languages

Co-authored-by: eglehelms <e.helms@cognigy.com>

src/api/types.ts

@@ -278,6 +278,10 @@ export interface SpeechCredential {
   custom_stt_endpoint: null | string;
   client_id: null | string;
   secret: null | string;
+  tts_api_key: null | string;
+  tts_region: null | string;
+  stt_api_key: null | string;
+  stt_region: null | string;
 }
 
 export interface Alert {

src/containers/internal/views/speech-services/form.tsx

@@ -25,6 +25,7 @@ import {
   VENDOR_NUANCE,
   VENDOR_WELLSAID,
   VENDOR_DEEPGRAM,
+  VENDOR_IBM,
 } from "src/vendor";
 import { MSG_REQUIRED_FIELDS } from "src/constants";
 import { getObscuredSecret } from "src/utils";
@@ -57,6 +58,10 @@ export const SpeechServiceForm = ({ credential }: SpeechServiceFormProps) => {
   const [secretKey, setSecretKey] = useState("");
   const [googleServiceKey, setGoogleServiceKey] =
     useState<GoogleServiceKey | null>(null);
+  const [sttRegion, setSttRegion] = useState("");
+  const [sttApiKey, setSttApiKey] = useState("");
+  const [ttsRegion, setTtsRegion] = useState("");
+  const [ttsApiKey, setTtsApiKey] = useState("");
   const [useCustomTts, setUseCustomTts] = useState(false);
   const [useCustomStt, setUseCustomStt] = useState(false);
   const [customTtsEndpoint, setCustomTtsEndpoint] = useState("");
@@ -111,6 +116,12 @@ export const SpeechServiceForm = ({ credential }: SpeechServiceFormProps) => {
           use_custom_stt: useCustomStt ? 1 : 0,
           custom_stt_endpoint: customSttEndpoint || null,
         }),
+        ...(vendor === VENDOR_IBM && {
+          stt_api_key: sttApiKey || null,
+          stt_region: sttRegion || null,
+          tts_api_key: ttsApiKey || null,
+          tts_region: ttsRegion || null,
+        }),
       };
 
       if (credential && credential.data) {
@@ -145,6 +156,10 @@ export const SpeechServiceForm = ({ credential }: SpeechServiceFormProps) => {
               : null,
           client_id: vendor === VENDOR_NUANCE ? clientId : null,
           secret: vendor === VENDOR_NUANCE ? secretKey : null,
+          stt_api_key: sttApiKey || null,
+          stt_region: sttRegion || null,
+          tts_api_key: ttsApiKey || null,
+          tts_region: ttsRegion || null,
         })
           .then(({ json }) => {
             toastSuccess("Speech credential created successfully");
@@ -211,6 +226,22 @@ export const SpeechServiceForm = ({ credential }: SpeechServiceFormProps) => {
         setSecretKey(credential.data.secret);
       }
 
+      if (credential.data.tts_api_key) {
+        setSttApiKey(credential.data.tts_api_key);
+      }
+
+      if (credential.data.tts_region) {
+        setTtsRegion(credential.data.tts_region);
+      }
+
+      if (credential.data.stt_api_key) {
+        setSttApiKey(credential.data.stt_api_key);
+      }
+
+      if (credential.data.stt_region) {
+        setSttRegion(credential.data.stt_region);
+      }
+
       setUseCustomTts(credential.data.use_custom_tts > 0 ? true : false);
       setUseCustomStt(credential.data.use_custom_stt > 0 ? true : false);
       setCustomTtsEndpoint(credential.data.custom_tts_endpoint || "");
@@ -404,27 +435,90 @@ export const SpeechServiceForm = ({ credential }: SpeechServiceFormProps) => {
             />
           </fieldset>
         )}
-        {/* Single region state var is used for both Microsoft and AWS */}
-        {regions && regions[vendor as keyof RegionVendors] && (
-          <fieldset>
-            <label htmlFor="region">
-              Region<span>*</span>
-            </label>
-            <Selector
-              id="region"
-              name="region"
-              value={region}
-              required
-              options={[
-                {
-                  name: "Select a region",
-                  value: "",
-                },
-              ].concat(regions[vendor as keyof RegionVendors])}
-              onChange={(e) => setRegion(e.target.value)}
-            />
-          </fieldset>
-        )}
+        {regions &&
+          regions[vendor as keyof RegionVendors] &&
+          vendor !== VENDOR_IBM && (
+            <fieldset>
+              <label htmlFor="region">
+                Region<span>*</span>
+              </label>
+              <Selector
+                id="region"
+                name="region"
+                value={region}
+                required
+                options={[
+                  {
+                    name: "Select a region",
+                    value: "",
+                  },
+                ].concat(regions[vendor as keyof RegionVendors])}
+                onChange={(e) => setRegion(e.target.value)}
+              />
+            </fieldset>
+          )}
+        {vendor === VENDOR_IBM &&
+          regions &&
+          regions[vendor as keyof RegionVendors] && (
+            <fieldset>
+              <label htmlFor="tts_region">
+                TTS Region {ttsCheck && <span>*</span>}
+              </label>
+              <Selector
+                id="tts_region"
+                name="tts_region"
+                value={ttsRegion}
+                required={ttsCheck}
+                options={[
+                  {
+                    name: "Select a region",
+                    value: "",
+                  },
+                ].concat(regions[vendor as keyof RegionVendors])}
+                onChange={(e) => setTtsRegion(e.target.value)}
+              />
+              <label htmlFor={`${vendor}_tts_apikey`}>
+                TTS API Key {ttsCheck && <span>*</span>}
+              </label>
+              <Passwd
+                id={`${vendor}_tts_apikey`}
+                required={ttsCheck}
+                name={`${vendor}_tts_apikey`}
+                placeholder="TTS API Key"
+                value={ttsApiKey ? getObscuredSecret(ttsApiKey) : ttsApiKey}
+                onChange={(e) => setTtsApiKey(e.target.value)}
+                disabled={credential ? true : false}
+              />
+              <label htmlFor="stt_region">
+                STT Region {sttCheck && <span>*</span>}
+              </label>
+              <Selector
+                id="stt_region"
+                name="stt_region"
+                value={sttRegion}
+                required={sttCheck}
+                options={[
+                  {
+                    name: "Select a region",
+                    value: "",
+                  },
+                ].concat(regions[vendor as keyof RegionVendors])}
+                onChange={(e) => setSttRegion(e.target.value)}
+              />
+              <label htmlFor={`${vendor}_sst_apikey`}>
+                SST API Key {sttCheck && <span>*</span>}
+              </label>
+              <Passwd
+                id={`${vendor}_stt_apikey`}
+                required={sttCheck}
+                name={`${vendor}_stt_apikey`}
+                placeholder="STT API Key"
+                value={sttApiKey ? getObscuredSecret(sttApiKey) : sttApiKey}
+                onChange={(e) => setSttApiKey(e.target.value)}
+                disabled={credential ? true : false}
+              />
+            </fieldset>
+          )}
         {vendor === VENDOR_MICROSOFT && (
           <React.Fragment>
             <fieldset>

src/vendor/index.tsx

@@ -15,6 +15,7 @@ export const VENDOR_MICROSOFT = "microsoft";
 export const VENDOR_WELLSAID = "wellsaid";
 export const VENDOR_NUANCE = "nuance";
 export const VENDOR_DEEPGRAM = "deepgram";
+export const VENDOR_IBM = "ibm";
 
 export const vendors: VendorOptions[] = [
   {
@@ -41,6 +42,10 @@ export const vendors: VendorOptions[] = [
     name: "deepgram",
     value: VENDOR_DEEPGRAM,
   },
+  {
+    name: "IBM",
+    value: VENDOR_IBM,
+  },
 ];
 
 export const useRegionVendors = () => {
@@ -52,14 +57,22 @@ export const useRegionVendors = () => {
     Promise.all([
       import("./regions/aws-regions"),
       import("./regions/ms-azure-regions"),
-    ]).then(([{ default: awsRegions }, { default: msRegions }]) => {
-      if (!ignore) {
-        setRegions({
-          aws: awsRegions,
-          microsoft: msRegions,
-        });
+      import("./regions/ibm-regions"),
+    ]).then(
+      ([
+        { default: awsRegions },
+        { default: msRegions },
+        { default: ibmRegions },
+      ]) => {
+        if (!ignore) {
+          setRegions({
+            aws: awsRegions,
+            microsoft: msRegions,
+            ibm: ibmRegions,
+          });
+        }
       }
-    });
+    );
 
     return function cleanup() {
       ignore = true;
@@ -84,11 +97,13 @@ export const useSpeechVendors = () => {
       import("./speech-recognizer/ms-speech-recognizer-lang"),
       import("./speech-recognizer/nuance-speech-recognizer-lang"),
       import("./speech-recognizer/deepgram-speech-recognizer-lang"),
+      import("./speech-recognizer/ibm-speech-recognizer-lang"),
       import("./speech-synthesis/aws-speech-synthesis-lang"),
       import("./speech-synthesis/google-speech-synthesis-lang"),
       import("./speech-synthesis/ms-speech-synthesis-lang"),
       import("./speech-synthesis/wellsaid-speech-synthesis-lang"),
       import("./speech-synthesis/nuance-speech-synthesis-lang"),
+      import("./speech-synthesis/ibm-speech-synthesis-lang"),
     ]).then(
       ([
         { default: awsRecognizer },
@@ -96,11 +111,13 @@ export const useSpeechVendors = () => {
         { default: msRecognizer },
         { default: nuanceRecognizer },
         { default: deepgramRecognizer },
+        { default: ibmRecognizer },
         { default: awsSynthesis },
         { default: googleSynthesis },
         { default: msSynthesis },
         { default: wellsaidSynthesis },
         { default: nuanceSynthesis },
+        { default: ibmSynthesis },
       ]) => {
         if (!ignore) {
           setSpeech({
@@ -110,6 +127,7 @@ export const useSpeechVendors = () => {
               microsoft: msSynthesis,
               wellsaid: wellsaidSynthesis,
               nuance: nuanceSynthesis,
+              ibm: ibmSynthesis,
             },
             recognizers: {
               aws: awsRecognizer,
@@ -117,6 +135,7 @@ export const useSpeechVendors = () => {
               microsoft: msRecognizer,
               nuance: nuanceRecognizer,
               deepgram: deepgramRecognizer,
+              ibm: ibmRecognizer,
             },
           });
         }

src/vendor/regions/ibm-regions.ts

@@ -0,0 +1,34 @@
+import type { Region } from "../types";
+
+export const regions: Region[] = [
+  {
+    name: "US South (Dallas)",
+    value: "us-south",
+  },
+  {
+    name: "US East (Washington)",
+    value: "us-east",
+  },
+  {
+    name: "EU DE (Frankfurt)",
+    value: "eu-de",
+  },
+  {
+    name: "EU GB (London)",
+    value: "eu-gb",
+  },
+  {
+    name: "AU South (Sydney)",
+    value: "au-syd",
+  },
+  {
+    name: "Japan (Tokyo)",
+    value: "jp-tok",
+  },
+  {
+    name: "South Korea (Seul)",
+    value: "kr-seo",
+  },
+];
+
+export default regions;

src/vendor/speech-recognizer/ibm-speech-recognizer-lang.ts

@@ -0,0 +1,86 @@
+import type { Language } from "../types";
+
+export const languages: Language[] = [
+  {
+    name: "Arabic (Modern Standard)",
+    code: "ar-MS_Telephony",
+  },
+  {
+    name: "Chinese (Mandarin)",
+    code: "zh-CN_Telephony",
+  },
+  {
+    name: "Czech ",
+    code: "cs-CZ_Telephony",
+  },
+  {
+    name: "Dutch (Belgian)",
+    code: "nl-BE_Telephony",
+  },
+  {
+    name: "Dutch (Netherlands)",
+    code: "nl-NL_Telephony",
+  },
+  {
+    name: "English (all supported dialects)",
+    code: "en-WW_Medical_Telephony",
+  },
+  {
+    name: "English (Australian)",
+    code: "en-AU_Telephony",
+  },
+  {
+    name: "English (Indian)",
+    code: "en-IN_Telephony",
+  },
+  {
+    name: "English (United Kingdom)",
+    code: "en-GB_Telephony",
+  },
+  {
+    name: "English (United States)",
+    code: "en-US_Telephony",
+  },
+  {
+    name: "French (Canadian)",
+    code: "fr-CA_Telephony",
+  },
+  {
+    name: "French (France)",
+    code: "fr-FR_Telephony",
+  },
+  {
+    name: "German",
+    code: "de-DE_Telephony",
+  },
+  {
+    name: "Hindi (Indian)",
+    code: "hi-IN_Telephony",
+  },
+  {
+    name: "Italian",
+    code: "it-IT_Telephony",
+  },
+  {
+    name: "Korean",
+    code: "ko-KR_Telephony",
+  },
+  {
+    name: "Portuguese (Brazilian)",
+    code: "pt-BR_Telephony",
+  },
+  {
+    name: "Spanish (Mexican)",
+    code: "es-LA_Telephony",
+  },
+  {
+    name: "Spanish (Castilian)",
+    code: "es-ES_Telephony",
+  },
+  {
+    name: "Swedish ",
+    code: "sv-SE_Telephony",
+  },
+];
+
+export default languages;

src/vendor/speech-synthesis/ibm-speech-synthesis-lang.ts

@@ -0,0 +1,171 @@
+import type { VoiceLanguage } from "../types";
+
+export const languages: VoiceLanguage[] = [
+  {
+    code: "de-DE",
+    name: "German (Germany)",
+    voices: [
+      { value: "de-DE_DieterVoice", name: "Dieter (Male): Standard German" },
+      {
+        value: "de-DE_DieterV2Voice",
+        name: "Dieter 2 (Male): Standard German",
+      },
+      {
+        value: "de-DE_DieterV3Voice",
+        name: "Dieter 3 (Male): Standard German",
+      },
+      { value: "de-DE_ErikaV3Voice", name: "Erika (Female): Standard German" },
+      { value: "de-DE_BirgitVoice", name: "Brigit (Female): Standard German" },
+      {
+        value: "de-DE_BirgitV2Voice",
+        name: "Brigit 2 (Female): Standard German",
+      },
+      {
+        value: "de-DE_BirgitV3Voice",
+        name: "Brigit 3 (Female): Standard German",
+      },
+    ],
+  },
+  {
+    code: "en-US",
+    name: "English (US)",
+    voices: [
+      {
+        value: "en-US_MichaelExpressive",
+        name: "Michael (Male): American English - Expressive",
+      },
+      { value: "en-US_MichaelVoice", name: "Michael (Male): American English" },
+      {
+        value: "en-US_MichaelV2Voice",
+        name: "Michael 2 (Male): American English",
+      },
+      {
+        value: "en-US_MichaelV3Voice",
+        name: "Michael 3 (Male): American English",
+      },
+      { value: "en-US_HenryV3Voice", name: "Henry (Male): American English" },
+      { value: "en-US_EmilyV3Voice", name: "Emily (Female): American English" },
+      {
+        value: "en-US_OliviaV3Voice",
+        name: "Olivia (Female): American English",
+      },
+      {
+        value: "en-US_AllisonExpressive",
+        name: "Allison (Female): American English - Expressive",
+      },
+      {
+        value: "en-US_AllisonVoice",
+        name: "Allison (Female): American English",
+      },
+      {
+        value: "en-US_AllisonV2Voice",
+        name: "Allison 2 (Female): American English",
+      },
+      {
+        value: "en-US_AllisonV3Voice",
+        name: "Allison 3 (Female): American English",
+      },
+      {
+        value: "en-US_LisaExpressive",
+        name: "Lisa (Female): American English - Expressive",
+      },
+      { value: "en-US_LisaVoice", name: "Lisa (Female): American English" },
+      { value: "en-US_LisaV2Voice", name: "Lisa 2 (Female): American English" },
+      { value: "en-US_LisaV3Voice", name: "Lisa 3 (Female): American English" },
+      { value: "en-US_KevinV3Voice", name: "Kevin (Male): American English" },
+      {
+        value: "en-US_EmmaExpressive",
+        name: "Emma (Female): American English - Expressive",
+      },
+    ],
+  },
+  {
+    code: "en-GB",
+    name: "English (GB)",
+    voices: [
+      { value: "en-GB_JamesV3Voice", name: "James (Male)" },
+      { value: "en-GB_KateVoice", name: "Kate (Female)" },
+      { value: "en-GB_KateV3Voice", name: "Kate 2 (Female)" },
+      { value: "en-GB_CharlotteV3Voice", name: "Kate (Female)" },
+    ],
+  },
+  {
+    code: "es-US",
+    name: "Spanish (North America)",
+    voices: [
+      {
+        value: "es-US_SofiaVoice",
+        name: "Sofia (Female): North American Spanish",
+      },
+      {
+        value: "es-US_SofiaV3Voice",
+        name: "Sofia 2 (Female): North American Spanish",
+      },
+    ],
+  },
+  {
+    code: "es-LA",
+    name: "Spanish (Latin America)",
+    voices: [
+      {
+        value: "es-LA_SofiaVoice",
+        name: "Sofia (Female): Latin American Spanish",
+      },
+      {
+        value: "es-LA_SofiaV3Voice",
+        name: "Sofia 2 (Female): Latin American Spanish",
+      },
+    ],
+  },
+  {
+    code: "es-ES",
+    name: "Spanish (Castilian)",
+    voices: [
+      { value: "es-ES_LauraVoice", name: "Laura (Female)" },
+      { value: "es-ES_LauraV3Voice", name: "Laura 2 (Female)" },
+      { value: "es-ES_EnriqueVoice", name: "Enrique (Male)" },
+      { value: "es-ES_EnriqueV3Voice", name: "Enrique 2 (Male)" },
+    ],
+  },
+  {
+    code: "fr-FR",
+    name: "French (FR)",
+    voices: [
+      { value: "fr-FR_NicolasV3Voice", name: "Nicolas (Male)" },
+      { value: "fr-FR_ReneeVoice", name: "Renee (Female)" },
+      { value: "fr-FR_ReneeV3Voice", name: "Renee 2 (Female)" },
+    ],
+  },
+  {
+    code: "fr-CA",
+    name: "French (CA)",
+    voices: [{ value: "fr-CA_LouiseV3Voice", name: "Louise (Female)" }],
+  },
+  {
+    code: "it-IT",
+    name: "Italian",
+    voices: [
+      { value: "it-IT_FrancescaVoice", name: "Francesca (Female)" },
+      { value: "it-IT_FrancescaV2Voice", name: "Francesca 2 (Female)" },
+      { value: "it-IT_FrancescaV3Voice", name: "Francesca 3 (Female)" },
+    ],
+  },
+  {
+    code: "pt-BR",
+    name: "Portuguese (Brazil)",
+    voices: [
+      { value: "pt-BR_IsabelaVoice", name: "Isabela (Female)" },
+      { value: "pt-BR_IsabelaV3Voice", name: "Isabela 2 (Female)" },
+    ],
+  },
+  {
+    code: "ja-JP",
+    name: "Japanese",
+    voices: [
+      { value: "ja-JP_EmiVoice", name: "Emi (Female)" },
+      { value: "ja-JP_EmiV3Voice", name: "Emi 2 (Female)" },
+    ],
+  },
+];
+
+export default languages;

src/vendor/types.ts

@@ -4,7 +4,8 @@ export type Vendor =
   | "Microsoft"
   | "WellSaid"
   | "Nuance"
-  | "deepgram";
+  | "deepgram"
+  | "IBM";
 
 export interface VendorOptions {
   name: Vendor;
@@ -49,6 +50,7 @@ export interface GoogleServiceKey {
 export interface RegionVendors {
   aws: Region[];
   microsoft: Region[];
+  ibm: Region[];
 }
 
 export interface RecognizerVendors {
@@ -57,6 +59,7 @@ export interface RecognizerVendors {
   microsoft: Language[];
   nuance: Language[];
   deepgram: Language[];
+  ibm: Language[];
 }
 
 export interface SynthesisVendors {
@@ -65,6 +68,7 @@ export interface SynthesisVendors {
   microsoft: VoiceLanguage[];
   wellsaid: VoiceLanguage[];
   nuance: VoiceLanguage[];
+  ibm: VoiceLanguage[];
 }
 
 export interface MSRawSpeech {