fix: do not parse URL path slashes as bare IPA

Bare IPA used /phoneme/ delimiters that overlapped with URL paths, so
segments like /foo/ were tokenized as IPA and plain-text or SSML output
dropped slashes.

Require a non host/path character before an opening slash so real bare
IPA still works after whitespace or punctuation. Add regression tests for
plain text and Alexa SSML.

Made-with: Cursor

Author: arjan

src/SpeechMarkdownGrammar.ts

@@ -269,7 +269,22 @@ export function speechMarkdownGrammar(myna: any): any {
       this.shortSubValue,
       '}',
     ).ast;
-    this.bareIpa = m.seq('/', this.shortIpaValue, '/').ast;
+    // Same characters as this.url / audio unquoted URLs (keep in sync with urlSpecialChar).
+    const urlUnquotedSpecials = ':/.-_~?#[]@!+,;%=()&';
+    // Bare IPA uses /phoneme/. The same delimiter appears in URL paths; only
+    // treat /.../ as IPA when '/' is not immediately after a host/path character
+    // (same set as letters, digits, and unquoted URL specials), e.g. after "foo+"
+    // or "https://example.com".
+    const bareIpaMayOpenHere = m.predicate((p: any) => {
+      if (p.index === 0) {
+        return true;
+      }
+      const prev = p.input.charAt(p.index - 1);
+      const urlInterior =
+        /[A-Za-z0-9]/.test(prev) || urlUnquotedSpecials.includes(prev);
+      return !urlInterior;
+    });
+    this.bareIpa = m.seq(bareIpaMayOpenHere, '/', this.shortIpaValue, '/').ast;
 
     const percentChange = ['+', m.hyphen, m.digit, '%'];
 
@@ -315,7 +330,7 @@ export function speechMarkdownGrammar(myna: any): any {
     ).ast;
 
     // Audio
-    this.urlSpecialChar = m.char(':/.-_~?#[]@!+,;%=()&');
+    this.urlSpecialChar = m.char(urlUnquotedSpecials);
     this.url = m.choice(m.digit, m.letter, this.urlSpecialChar).oneOrMore.ast;
     this.audio = m.seq(
       '!',

tests/ipa-short.spec.ts

@@ -75,4 +75,25 @@ describe('ipa-short', () => {
     `;
     expect(text).toBe(expectedText);
   });
+ 
+  test('plain text keeps URL path slashes (does not treat as bare IPA)', () => {
+    const input = 'see https://example.com/foo/bar';
+    expect(speech.toText(input, {})).toBe(input);
+    expect(speech.toText('https://a.com/x/y', {})).toBe('https://a.com/x/y');
+    expect(speech.toText('see example.com/foo/bar', {})).toBe(
+      'see example.com/foo/bar',
+    );
+    // '+' and other unquoted URL specials before '/' must not open bare IPA.
+    expect(speech.toText('https://x.com/seg+/y', {})).toBe(
+      'https://x.com/seg+/y',
+    );
+  });
+
+  test('SSML keeps URL path slashes (does not treat as bare IPA)', () => {
+    const ssml = speech.toSSML('see https://example.com/foo/bar', {
+      platform: 'amazon-alexa',
+    });
+    expect(ssml).toContain('https://example.com/foo/bar');
+    expect(ssml).not.toContain('<phoneme');
+  });
 });